copilot-custom-endpoint 3.6.0 → 3.7.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -20,20 +20,17 @@ That's it. No code, no servers to manage (unless the model specifically needs th
20
20
 
21
21
  ## Pick a model
22
22
 
23
- | Model | Provider | Needs proxy? | Vision | Setup guide |
24
- | ---------------------------- | ---------- | ----------------------- | ------ | ----------------------------------------------------------------------------------------------------------------------------------------- |
25
- | **Kimi K3.1 Pro** | Moonshot | **Yes** | ✅ | [Manual setup](docs/models/kimi.md) |
26
- | **Kimi K3** | Moonshot | **Yes** | ✅ | [Manual setup](docs/models/kimi.md) |
27
- | **MiMo V2.6 Pro** | Xiaomi | **Yes** | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=sdmapvstool.xiaomimimo-for-copilot) / [Manual setup](docs/models/mimo.md) |
28
- | **MiMo V2.6 Flash** | Xiaomi | **Yes** | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=sdmapvstool.xiaomimimo-for-copilot) / [Manual setup](docs/models/mimo.md) |
29
- | **MiMo V2.6 Pro UltraSpeed** | Xiaomi | **Yes** | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=sdmapvstool.xiaomimimo-for-copilot) / [Manual setup](docs/models/mimo.md) |
30
- | **Qwen 3.8 Max (0803)** | OpenRouter | Optional (experimental) | ✅ | [Snapshot/proxy setup](docs/models/qwen.md#optional-openrouter-snapshot-proxy) |
31
- | **Qwen 3.8 Max (0902)** | DashScope | Optional (recommended) | ✅ | [Manual setup](docs/models/qwen.md) |
32
- | **MiniMax M3.1 Flash** | MiniMax | No | ✅ | [Manual setup](docs/models/minimax.md) |
33
- | **MiniMax M3** | MiniMax | No | ✅ | [Extension](https://github.com/tugudush/minimax-copilot) ★ / [Manual setup](docs/models/minimax.md) |
34
- | **GLM 5.3 Flash** | Z.ai | No | ✅ | [Manual setup](docs/models/glm.md) |
35
- | **GLM 5.3** | Z.ai | No | ❌ | [Manual setup](docs/models/glm.md) |
36
- | **DeepSeek V4.1 Flash 1017** | DeepSeek | No (uses an extension) | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=Vizards.deepseek-v4-for-copilot) / [setup notes](docs/models/deepseek.md) |
23
+ | Model | Provider | Needs proxy? | Vision | Setup guide |
24
+ | ---------------------------- | --------- | ---------------------- | ------ | ----------------------------------------------------------------------------------------------------------------------------------------- |
25
+ | **Kimi K3.1 Pro** | Moonshot | **Yes** | ✅ | [Manual setup](docs/models/kimi.md) |
26
+ | **MiMo V2.6 Pro** | Xiaomi | **Yes** | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=sdmapvstool.xiaomimimo-for-copilot) / [Manual setup](docs/models/mimo.md) |
27
+ | **MiMo V2.6 Flash** | Xiaomi | **Yes** | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=sdmapvstool.xiaomimimo-for-copilot) / [Manual setup](docs/models/mimo.md) |
28
+ | **MiMo V2.6 Pro UltraSpeed** | Xiaomi | **Yes** | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=sdmapvstool.xiaomimimo-for-copilot) / [Manual setup](docs/models/mimo.md) |
29
+ | **Qwen 3.8 Max** | DashScope | Optional (recommended) | ✅ | [Manual setup](docs/models/qwen.md) |
30
+ | **MiniMax M3.1 Flash** | MiniMax | No | ✅ | [Manual setup](docs/models/minimax.md) |
31
+ | **GLM 5.3 Flash** | Z.ai | No | ✅ | [Manual setup](docs/models/glm.md) |
32
+ | **GLM 5.3** | Z.ai | No | ❌ | [Manual setup](docs/models/glm.md) |
33
+ | **DeepSeek V4.1 Flash 1017** | DeepSeek | No (uses an extension) | ✅ | [Extension](https://marketplace.visualstudio.com/items?itemName=Vizards.deepseek-v4-for-copilot) / [setup notes](docs/models/deepseek.md) |
37
34
 
38
35
  ## Setup
39
36
 
@@ -122,7 +119,19 @@ It approximates the cost of one point on the AA Intelligence Index under the com
122
119
 
123
120
  For example, GLM 5.3 Flash is `(0.5 × $0.15) + (0.1 × $0.50) = $0.125`, shown as `~$0.13`. Cached-input discounts are not included. Actual costs vary with context length, output length, caching, pricing tiers, promotions, and provider billing. DeepSeek estimates use peak rates; official off-peak rates are half.
124
121
 
125
- The current AA values are **Artificial Analysis Intelligence Index v5.0** composites taken from OpenRouter’s model metadata; this snapshot was re-based from v4.3.2 on October 31, 2026, so its scores are not comparable with earlier revisions of this table. **Every pricing cell comes from first-party provider pricing or GitHub Copilot’s official billing table, never OpenRouter.** AA’s current leaderboard no longer publishes a v5.0 composite for nine older rows — GPT-5.6 Luna, Qwen 3.8 Max (0803), Grok 4.6, GPT-6 Sol, Claude Sonnet 5, GPT-5.6 Terra, GPT-5.6 Sol, Claude Opus 5, and Claude Fable 5 — so their Cost per intelligence shows `—` and they are listed after the scored rows by estimated session cost. The Qwen ranking-card figure of **53.4 (#3)** is a v4.3.2 result; the exact 0902 metadata row is **33.6** under v5.0, and Qwen Cloud PAYG pricing remains **$2 / $0.25 implicit cache / $6**. This snapshot records Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.7, GPT-6 Sol, GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5 as **Copilot-native** models. **GPT-6.1 Terra** (released October 2026, in Copilot since October 26) enters at **51** on flat pricing of **$1 / $0.05 cache read / $5** per 1M tokens with vision and a 1M context, which puts it at **~$0.0196** per intelligence point — about half of GPT-6.1 Sol’s **~$0.0367**. Gemini 4 Argon (released October 13, 2026) scores **50.4** on Google’s introductory pricing of **$2 / $0.10 implicit cache / $10** per 1M tokens, which doubles to **$4 / $0.20 / $20** on December 1, 2026. Claude Haiku 5.5 (released October 7, 2026) scores **36.2** and, at **~$0.0028** per intelligence point, is still the lowest cost per intelligence in the table; MiMo V2.6 Flash (**26.7**) is still the cheapest custom-endpoint option at **~$0.0037**. Muse Spark 1.3 (Meta, released September 2, 2026) scores **36.4** on Meta Model API pricing of **$1.25 / $0.15 cached input / $4.25** per 1M tokens; its day-one `muse-spark-1.3-contributor` SKU is excluded from this snapshot because Meta may train future models on those prompts and completions. DeepSeek’s **V4.1-Flash-1017** build (released October 17, 2026) scores **32** at unchanged peak rates ($0.30 / $0.006 / $1.20), or **~$0.0084** per intelligence point. MiniMax’s **MiniMax-M3.1-Flash** (announced October 19, 2026) scores **28.9** on flat pricing of **$0.15 / $0.015 cache read / $0.60** per 1M tokens — no prompt-length tiers and no `service_tier` options — which puts it at **~$0.0047** per intelligence point. Kimi K3.1 Pro (**46**) and GPT-6.1 Sol (**54.5**) remain served at their existing rates.
122
+ The current AA values are **Artificial Analysis Intelligence Index v5.0** composites taken from OpenRouter’s model metadata; this snapshot was re-based from v4.3.2 on October 31, 2026, so its scores are not comparable with earlier revisions of this table.
123
+
124
+ **Every pricing cell comes from first-party provider pricing or GitHub Copilot’s official billing table, never OpenRouter.** The Qwen 3.8 Max row scores **33.6** under v5.0, and Qwen Cloud PAYG pricing remains **$2 / $0.25 implicit cache / $6**.
125
+
126
+ This snapshot records Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.7, GPT-6 Sol, GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna, Claude Fable 5.2, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5 as **Copilot-native** models.
127
+
128
+ **GPT-6.1 Terra** (released October 2026, in Copilot since October 26) enters at **51** on flat pricing of **$1 / $0.05 cache read / $5** per 1M tokens with vision and a 1M context, which puts it at **~$0.0196** per intelligence point — about half of GPT-6.1 Sol’s **~$0.0367**. Gemini 4 Argon (released October 13, 2026) scores **50.4** on Google’s introductory pricing of **$2 / $0.10 implicit cache / $10** per 1M tokens, which doubles to **$4 / $0.20 / $20** on December 1, 2026.
129
+
130
+ Claude Haiku 5.5 (released October 7, 2026) scores **36.2** and, at **~$0.0028** per intelligence point, is still the lowest cost per intelligence in the table; MiMo V2.6 Flash (**26.7**) is still the cheapest custom-endpoint option at **~$0.0037**.
131
+
132
+ Muse Spark 1.3 (Meta, released September 2, 2026) scores **36.4** on Meta Model API pricing of **$1.25 / $0.15 cached input / $4.25** per 1M tokens; its day-one `muse-spark-1.3-contributor` SKU is excluded from this snapshot because Meta may train future models on those prompts and completions.
133
+
134
+ DeepSeek’s **V4.1-Flash-1017** build (released October 17, 2026) scores **32** at unchanged peak rates ($0.30 / $0.006 / $1.20), or **~$0.0084** per intelligence point. MiniMax’s **MiniMax-M3.1-Flash** (announced October 19, 2026) scores **28.9** on flat pricing of **$0.15 / $0.015 cache read / $0.60** per 1M tokens — no prompt-length tiers and no `service_tier` options — which puts it at **~$0.0047** per intelligence point. Kimi K3.1 Pro (**46**) and GPT-6.1 Sol (**54.5**) remain served at their existing rates.
126
135
 
127
136
  | Model | Provider | Cost per intelligence | Intelligence Score | Est. session | Vision | Context window |
128
137
  | ---------------------------- | --------- | --------------------- | ------------------ | ------------ | ------ | -------------- |
@@ -133,7 +142,6 @@ The current AA values are **Artificial Analysis Intelligence Index v5.0** compos
133
142
  | **MiniMax M3.1 Flash** | MiniMax | **~$0.0047** | **28.9** | ~$0.14 | ✅ | 1M |
134
143
  | **DeepSeek V4.1 Flash 1017** | DeepSeek | **~$0.0084** | **32** | ~$0.27 | ✅ | 1M |
135
144
  | **MiMo V2.6 Pro** | Xiaomi | **~$0.0102** | **29.9** | ~$0.30 | ✅ | 1M |
136
- | **MiniMax M3** | MiniMax | **~$0.0144** | **18.7** | ~$0.27 | ✅ | 1M |
137
145
  | **GPT-6.1 Terra** | OpenAI | **~$0.0196** | **51** | ~$1.00 | ✅ | 1M |
138
146
  | **Gemini 3.8 Flash** | Google | **~$0.0246** | **30.5** | ~$0.75 | ✅ | 1M |
139
147
  | **Muse Spark 1.3** | Meta | **~$0.0288** | **36.4** | ~$1.05 | ✅ | 1M |
@@ -142,35 +150,27 @@ The current AA values are **Artificial Analysis Intelligence Index v5.0** compos
142
150
  | **Claude Sonnet 5.5** | Anthropic | **~$0.0369** | **54.2** | ~$2.00 | ✅ | 1M |
143
151
  | **Gemini 4 Argon** | Google | **~$0.0397** | **50.4** | ~$2.00 | ✅ | 1M |
144
152
  | **Kimi K3.1 Pro** | Moonshot | **~$0.0435** | **46** | ~$2.00 | ✅ | 1M |
153
+ | **GPT-6 Sol** | OpenAI | **~$0.0455** | **44** | ~$2.00 | ✅ | 1M |
145
154
  | **Grok 4.7** | xAI | **~$0.0457** | **35** | ~$1.60 | ✅ | 500K |
146
- | **Qwen 3.8 Max (0902)** | DashScope | **~$0.0476** | **33.6** | ~$1.60 | ✅ | 1M |
155
+ | **Qwen 3.8 Max** | DashScope | **~$0.0476** | **33.6** | ~$1.60 | ✅ | 1M |
147
156
  | **Claude Opus 5.5** | Anthropic | **~$0.0685** | **58.4** | ~$4.00 | ✅ | 1M |
148
157
  | **MiMo V2.6 Pro UltraSpeed** | Xiaomi | **~$0.1018** | **29.9** | ~$3.05 | ✅ | 1M |
149
- | **Kimi K3** | Moonshot | **~$0.1724** | **17.4** | ~$3.00 | ✅ | 1M |
158
+ | **Claude Fable 5.2** | Anthropic | **~$0.1231** | **65** | ~$8.00 | ✅ | 1M |
150
159
  | **GPT-6 Astra** | OpenAI | **~$0.1757** | **56.9** | ~$10.00 | ✅ | 1M |
151
160
  | **Claude Fable 5.1** | Anthropic | **~$0.2012** | **49.7** | ~$10.00 | ✅ | 1M |
152
- | **GPT-5.6 Luna** | OpenAI | — | — | ~$0.22 | ✅ | 1M |
153
- | **Qwen 3.8 Max (0803)** | DashScope | — | — | ~$1.60 | ✅ | 1M |
154
- | **Grok 4.6** | xAI | — | — | ~$1.60 | ✅ | 500K |
155
- | **GPT-6 Sol** | OpenAI | — | — | ~$2.00 | ✅ | 1M |
156
- | **Claude Sonnet 5** | Anthropic | — | — | ~$2.00 | ✅ | 1M |
157
- | **GPT-5.6 Terra** | OpenAI | — | — | ~$2.20 | ✅ | 1M |
158
- | **GPT-5.6 Sol** | OpenAI | — | — | ~$4.00 | ✅ | 1M |
159
- | **Claude Opus 5** | Anthropic | — | — | ~$5.00 | ✅ | 1M |
160
- | **Claude Fable 5** | Anthropic | — | — | ~$10.00 | ✅ | 1M |
161
-
162
- > `MiMo V2.6 Pro UltraSpeed` uses the MiMo V2.6 Pro benchmark score because it is the same-quality latency tier; no independent composite is published. `MiniMax M3.1 Flash` is a separate model with its own flat rates; MiniMax's API rejects `service_tier` for it. `Kimi K3.1 Pro` supersedes K3 at a lower price ($2 / $0.20 / $10 vs $3 / $0.30 / $15 per 1M); K3 remains served. `MiMo V2.6 Flash`'s composite (26.7) is OpenRouter's Artificial Analysis v5.0 value; Xiaomi's published launch benchmarks for it are recorded in [docs/benchmarks.md](docs/benchmarks.md).
163
- > GPT-6 Sol, GPT-6.1 Sol, and Luna use standard rates up to 272K input tokens; requests above that threshold have higher input/cache and output rates. GitHub Copilot also bills cache writes separately. Plan access differs: GPT-6 Sol and GPT-6.1 Sol require Pro+ or higher, Luna and Claude Sonnet 5.5 are available from Pro, Claude Opus 5.5 requires Pro+ or higher, and Claude Haiku 5.5 is available on every paid plan (Pro through Enterprise). Gemini 4 Argon is Copilot-native for Pro+, Max, Business, and Enterprise (not Pro); Google's $2 / $0.10 implicit cache / $10 per 1M introductory pricing (output includes thinking tokens, implicit caching only) is guaranteed only through November 30, 2026 and doubles to $4 / $0.20 / $20 on December 1, 2026.
164
- > Claude Haiku 5.5 is the only current Claude model priced by prompt length: $0.10 / $0.01 cache read / $0.50 per 1M for prompts up to 100,000 tokens, and $0.50 / $0.05 / $2.50 above that. The table shows the sub-100K rates, which Anthropic says cover ~90% of Haiku-class requests.
165
- > Muse Spark 1.3 is closed and API-only on the Meta Model API (`muse-spark-1.3`); the 12–21× cheaper `muse-spark-1.3-contributor` SKU is excluded from this snapshot because Meta may use prompts and completions to train future models. Meta reports audio understanding is degraded versus 1.2.
161
+
162
+ > `MiMo V2.6 Pro UltraSpeed` uses the MiMo V2.6 Pro score (same quality, lower latency).
163
+ > Plan access: GPT-6 Sol, GPT-6.1 Sol, Claude Opus 5.5, Claude Fable 5.2, and Gemini 4 Argon need Pro+ or higher; GPT-6 Luna and Claude Sonnet 5.5 are available from Pro; Claude Haiku 5.5 is on every paid plan.
164
+ > Claude Haiku 5.5 is priced by prompt length; the table shows its sub-100K rates.
166
165
 
167
166
  > Detailed pricing, benchmark provenance, and model-specific notes live in [docs/pricing.md](docs/pricing.md) and [docs/benchmarks.md](docs/benchmarks.md). This page stays focused on setup and model selection. For a copy-paste config containing **all providers at once**, see [docs/example-config.md](docs/example-config.md).
168
167
 
169
168
  > **👤 Personal picks** —
170
169
  >
171
170
  > - **Claude Haiku 5.5** — the lowest cost per intelligence in this table (~$0.0028) and the best value among Copilot-native models, at about $0.10/session with vision.
172
- > - **MiMo V2.6 Flash** — the lowest cost per intelligence among custom-endpoint models (~$0.0037), at about $0.10/session with vision.
173
- > - **GLM 5.3 Flash** — the runner-up among custom-endpoint models (~$0.0045), at about $0.13/session with vision.
171
+ > - **GPT-6.1 Terra** — a high score (**51**) at a low cost per intelligence (~$0.0196), about $1.00/session with vision.
172
+ > - **GLM 5.3 Flash** — the runner-up among custom-endpoint models (~$0.0045), at about $0.13/session with vision. Well reviewed, alongside DeepSeek V4.1 Flash.
173
+ > - **DeepSeek V4.1 Flash 1017** — well reviewed, alongside GLM 5.3 Flash; about $0.27/session with vision (~$0.0084 per intelligence point at peak rates).
174
174
 
175
175
  ## Companion tools
176
176
 
@@ -182,25 +182,12 @@ These work alongside the providers above and fill gaps that VS Code's built-in t
182
182
 
183
183
  VS Code's built-in `view_image` tool only accepts **static images** (PNG, JPG, GIF, WebP). That's a hard wall if you want to ask an AI assistant about a screen recording, a screencast, a product demo, or any other video. Several vision-capable models in this repo actually accept video natively — but VS Code's tool pipeline never gets the chance to forward it.
184
184
 
185
- **Video Context MCP** is a small MCP server that bridges that gap. It works with **GitHub Copilot, Cursor, and Claude Code** out of the box, and:
186
-
187
- - **Extracts frames** from local files or remote URLs (no `ffmpeg` gymnastics required).
188
- - **Routes them through a multi-provider fallback chain** — `Gemini → GLM 4.6V Flash → Qwen3.8-max`.
189
- - **Answers natural-language questions** about the video grounded in actual frames: "what does the speaker click in the last 30 seconds?", "summarize the demo", "find the frame where the error appears".
190
- - **Extras:** timestamp search, audio transcription with speaker diarization, and video metadata (resolution, duration, codec).
191
-
192
- ### 🪣 [Bitbucket MCP](https://bitbucketmcp.tugudush.com/) — _secure, read-only Bitbucket access for VS Code Copilot, Cursor, and Claude Code_
193
-
194
- GitHub ships a first-party MCP server (and it's even bundled into Copilot), so asking "what's open in my org's repos / show me PR #123" works seamlessly on github.com. **Bitbucket has no equivalent** — Atlassian hasn't shipped one — which leaves Bitbucket Cloud users copy-pasting PR URLs, diffs, and file contents into chat by hand.
195
-
196
- **Bitbucket MCP** is a small MCP server that closes that gap. It works with **VS Code GitHub Copilot, Cursor, and Claude Code** out of the box, and:
185
+ **Video Context MCP** is a small MCP server that bridges that gap. It works with **GitHub Copilot, Cursor, Claude Code, OpenAI Codex, and Google Antigravity** out of the box, and:
197
186
 
198
- - **38 tools across 8 categories** — repositories, pull requests, branches & commits, diffs & comparisons, CI/CD pipelines, issues, code search, and users.
199
- - **Read-only by design** — `makeRequest()` blocks all non-GET requests at runtime, so no write, delete, or modify operation is possible.
200
- - **Token-friendly output** — every tool supports `text`, `json`, and `toon` (Token-Oriented Object Notation) formats; `toon` cuts LLM token consumption by 30–60% on large PR/commit lists.
201
- - **JMESPath filtering** on all 38 tools, so you can trim responses (e.g. only open PRs, or just title + author) before they hit the model.
202
- - **One-call PR context** — `bb_get_context` bundles PR metadata, diffstat, CI statuses, and comments in a single request.
203
- - **Drop-in install** — `npm install -g @tugudush/bitbucket-mcp` plus a short `.vscode/mcp.json` entry, authenticated with a Bitbucket API token + your Atlassian email.
187
+ - **Answers natural-language questions** about the video grounded in actual frames and the audio transcript: "what does the speaker click in the last 30 seconds?", "find the frame where the error appears". It can also generate structured summaries with key scenes and timelines.
188
+ - **Extracts frames** at specific timestamps, scene changes, or fixed intervals, from local files or remote `http(s)` URLs.
189
+ - **Falls back across providers** — Gemini 3.8 Flash (default) → MiniMax-M3.1-Flash → Kimi K3.1 Pro → Qwen3.8 Max → MiMo-V2.6 → GLM-5.3-Flash. If a key is missing or a call fails, the next provider is tried.
190
+ - **Extras:** timestamp search, audio transcription with speaker diarization and SRT/VTT/JSON export, and video metadata (resolution, duration, fps, codec).
204
191
 
205
192
  ## Need help?
206
193
 
@@ -223,7 +210,7 @@ GitHub ships a first-party MCP server (and it's even bundled into Copilot), so a
223
210
  2. If the provider needs request rewriting, add a proxy under `proxy/`.
224
211
  3. Submit a PR.
225
212
 
226
- The current AA values are **Artificial Analysis Intelligence Index v5.0** composites taken from OpenRouter’s model metadata (re-based from v4.3.2 on October 31, 2026); **every pricing cell comes from first-party provider pricing or GitHub Copilot’s official billing table, never OpenRouter.** Nine older rows have no v5.0 composite on AA’s current leaderboard and therefore show `—` for Cost per intelligence. Qwen Cloud PAYG pricing remains **$2 / $0.25 implicit cache / $6** for Qwen3.8 Max; its separate Token Plan is not the PAYG rate table. This snapshot records Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.7, GPT-6 Sol, GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5 as **Copilot-native** models.
213
+ The current AA values are **Artificial Analysis Intelligence Index v5.0** composites taken from OpenRouter’s model metadata (re-based from v4.3.2 on October 31, 2026); **every pricing cell comes from first-party provider pricing or GitHub Copilot’s official billing table, never OpenRouter.** Qwen Cloud PAYG pricing remains **$2 / $0.25 implicit cache / $6** for Qwen3.8 Max; its separate Token Plan is not the PAYG rate table. This snapshot records Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.7, GPT-6 Sol, GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna, Claude Fable 5.2, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5 as **Copilot-native** models.
227
214
 
228
215
  - **Chat only.** Inline completions, semantic search, and next-edit suggestions still need a GitHub-hosted model.
229
216
  - Each proxy is tuned for a specific provider family. Don't point the Kimi proxy at an arbitrary OpenAI-compatible endpoint.
@@ -6,7 +6,7 @@ Here's a complete, real-world `chatLanguageModels.json` that combines **the `cus
6
6
  >
7
7
  > **Note:** The `apiKey` fields are left as empty strings — set them via the **Chat: Manage Language Models** UI (Command Palette → right-click provider group → **Update API Key**). After you set a key via the UI, VS Code replaces the empty string with a `${input:chat.lm.secret.<id>}` secret reference.
8
8
  >
9
- > The live config points Qwen at the local proxy (`:3458`) and MiMo at the local proxy (`:3459`). When using a proxy, align the model `requestBody` overrides with the proxy's behavior: Qwen sends no `requestBody` (the proxy manages `enable_thinking` dynamically); MiMo sends only `temperature` and `top_p` (the proxy injects `thinking: {"type": "disabled"}` when tool history is missing `reasoning_content` and leaves it absent on plain chat). OpenRouter lists Qwen 3.8 Max as two frozen snapshots (`qwen/qwen3.8-max` = 0803 and `qwen/qwen3.8-max-0902` = 0902); this DashScope config intentionally uses the provider ID `qwen3.8-max` and does not pretend the OpenRouter slugs are interchangeable.
9
+ > The live config points Qwen at the local proxy (`:3458`) and MiMo at the local proxy (`:3459`). When using a proxy, align the model `requestBody` overrides with the proxy's behavior: Qwen sends no `requestBody` (the proxy manages `enable_thinking` dynamically); MiMo sends only `temperature` and `top_p` (the proxy injects `thinking: {"type": "disabled"}` when tool history is missing `reasoning_content` and leaves it absent on plain chat). This DashScope config uses the provider ID `qwen3.8-max`.
10
10
 
11
11
  ```json
12
12
  [
@@ -47,20 +47,6 @@ Here's a complete, real-world `chatLanguageModels.json` that combines **the `cus
47
47
  "streaming": true,
48
48
  "maxInputTokens": 1000000,
49
49
  "maxOutputTokens": 131072
50
- },
51
- {
52
- "id": "kimi-k3",
53
- "name": "Kimi K3 (vision)",
54
- "url": "http://127.0.0.1:3457/v1/chat/completions",
55
- "requestBody": {
56
- "temperature": 1,
57
- "max_tokens": 8192
58
- },
59
- "toolCalling": true,
60
- "vision": true,
61
- "streaming": true,
62
- "maxInputTokens": 1000000,
63
- "maxOutputTokens": 131072
64
50
  }
65
51
  ]
66
52
  },
@@ -136,22 +122,6 @@ Here's a complete, real-world `chatLanguageModels.json` that combines **the `cus
136
122
  "temperature": 1,
137
123
  "top_p": 0.95
138
124
  }
139
- },
140
- {
141
- "id": "MiniMax-M3",
142
- "name": "MiniMax M3 (vision)",
143
- "url": "https://api.minimax.io/v1/chat/completions",
144
- "toolCalling": true,
145
- "vision": true,
146
- "streaming": true,
147
- "maxInputTokens": 1048576,
148
- "maxOutputTokens": 131072,
149
- "requestBody": {
150
- "thinking": { "type": "adaptive" },
151
- "reasoning_split": true,
152
- "temperature": 1,
153
- "top_p": 0.95
154
- }
155
125
  }
156
126
  ]
157
127
  },
@@ -202,10 +172,10 @@ Here's a complete, real-world `chatLanguageModels.json` that combines **the `cus
202
172
 
203
173
  If you only need one provider, jump straight to its setup guide:
204
174
 
205
- - [Kimi K3 / K2.6 / K2.7 Code](kimi.md)
175
+ - [Kimi K3.1 Pro / K2.6 / K2.7 Code](kimi.md)
206
176
  - [Qwen 3.8 Max](qwen.md)
207
177
  - [Xiaomi MiMo (V2.6 Pro / Flash / Pro UltraSpeed)](mimo.md)
208
- - [MiniMax M3.1 Flash / M3](minimax.md)
178
+ - [MiniMax M3.1 Flash](minimax.md)
209
179
  - [GLM (5.3 Flash / 5.3 / 5.2 / 5.1)](glm.md)
210
180
 
211
181
  > **DeepSeek V4.1 Flash 1017** uses the [DeepSeek for Copilot Chat extension](https://marketplace.visualstudio.com/items?itemName=Vizards.deepseek-v4-for-copilot). It appears in `chatLanguageModels.json` as `vendor: "deepseek"` (not as a `customendpoint` provider) and is configured via the extension's settings block.
@@ -1,10 +1,8 @@
1
1
  # Kimi (Moonshot) — VS Code Custom Endpoint Setup Guide
2
2
 
3
- > **TL;DR:** The Kimi K3 family requires the local proxy. These models are always-thinking and use `reasoning_effort` (not the `thinking` parameter) while the proxy enforces Kimi's fixed sampling constraints. Direct VS Code → Moonshot is not viable.
3
+ > **TL;DR:** Kimi K3.1 Pro requires the local proxy. It is always-thinking and uses `reasoning_effort` (not the `thinking` parameter) while the proxy enforces Kimi's fixed sampling constraints. Direct VS Code → Moonshot is not viable.
4
4
  >
5
- > **🆕 Kimi K3.1 Pro** (released October 23, 2026) is the current flagship — a post-trained successor to K3 (3T class, 1M context) with AA Intelligence Index **55.1**, ahead of Claude Fable 5.1 and behind Claude Sonnet 5.5 in our roster. Always-thinking with `reasoning_effort` `low` / `high` / `max` (default `max`). Open weights promised by November 4, 2026.
6
- >
7
- > **Previous:** **Kimi K3** (July 16, 2026) — 2.8T parameters, 1M context, AA Index **43.6**. Still served and still supported by the proxy; use K3.1 Pro unless you specifically need K3.
5
+ > **🆕 Kimi K3.1 Pro** (released October 23, 2026) is the current flagship — a 3T-class model with 1M context and AA Intelligence Index **55.1**, ahead of Claude Fable 5.1 and behind Claude Sonnet 5.5 in our roster. Always-thinking with `reasoning_effort` `low` / `high` / `max` (default `max`). Open weights promised by November 4, 2026.
8
6
  >
9
7
  > **🆕 Alternative:** On the **Kimi Coding tier** (`kimi.com/code/console`), the [**Moonshot LM Provider**](https://marketplace.visualstudio.com/items?itemName=DenizhanDaklr.kimi-lm-provider) extension offers zero-proxy setup with reasoning visibility. **Not compatible with Pay-as-You-Go** keys from `platform.kimi.ai`. See [the extension comparison](../research/kimi-vscode-extension.md).
10
8
 
@@ -24,13 +22,12 @@
24
22
 
25
23
  ### Models
26
24
 
27
- | Model | Vision | Context | Max output | Notes |
28
- | ------------------------- | ------ | ------- | --------------- | ---------------------------------------------------------------------------------------------------------------------- |
29
- | `kimi-k3.1-pro` | ✅ Yes | 1M | 131072 (1M max) | **Flagship.** Post-trained K3, 3T class. Always-thinking, uses `reasoning_effort` (not `thinking`). AA Index **55.1**. |
30
- | `kimi-k3.1-pro-highspeed` | ✅ Yes | 1M | 131072 | 2–3× faster SKU of K3.1 Pro; same capabilities at a higher per-token price. |
31
- | `kimi-k3` | ✅ Yes | 1M | 131072 | Previous flagship. 2.8T params. AA Index **43.6**. |
25
+ | Model | Vision | Context | Max output | Notes |
26
+ | ------------------------- | ------ | ------- | --------------- | ----------------------------------------------------------------------------------------------------- |
27
+ | `kimi-k3.1-pro` | ✅ Yes | 1M | 131072 (1M max) | **Flagship.** 3T class. Always-thinking, uses `reasoning_effort` (not `thinking`). AA Index **55.1**. |
28
+ | `kimi-k3.1-pro-highspeed` | ✅ Yes | 1M | 131072 | 2–3× faster SKU of K3.1 Pro; same capabilities at a higher per-token price. |
32
29
 
33
- > This guide documents the K3.1 Pro setup. K3.1 Pro and K3 share the same proxy and request shape, so the snippet below works for both — change only the `id`. Deprecated K2 ids (`kimi-k2-0905-preview`, `kimi-k2-turbo-preview`, `kimi-k2-thinking`, etc.) were discontinued May 25, 2026, and Moonshot plans to retire the remaining K2 series by the end of December 2026.
30
+ > This guide documents the K3.1 Pro setup. Deprecated K2 ids (`kimi-k2-0905-preview`, `kimi-k2-turbo-preview`, `kimi-k2-thinking`, etc.) were discontinued May 25, 2026, and Moonshot plans to retire the remaining K2 series by the end of December 2026.
34
31
 
35
32
  ## Quick Start
36
33
 
@@ -67,25 +64,12 @@
67
64
  "streaming": true,
68
65
  "maxInputTokens": 1000000,
69
66
  "maxOutputTokens": 131072
70
- },
71
- {
72
- "id": "kimi-k3",
73
- "name": "Kimi K3 (Custom Endpoint)",
74
- "url": "http://127.0.0.1:3457/v1/chat/completions",
75
- "requestBody": { "temperature": 1, "max_tokens": 8192 },
76
- "toolCalling": true,
77
- "vision": true,
78
- "streaming": true,
79
- "maxInputTokens": 1000000,
80
- "maxOutputTokens": 131072
81
67
  }
82
68
  ]
83
69
  }
84
70
  ```
85
71
 
86
72
  > **K3.1 Pro note:** `max_tokens` is set to **8192** as a safe starting point. K3.1 Pro is always-thinking and very verbose (150M tokens generated on the AA Intelligence Index evaluation versus an 87M median). The API default for `max_completion_tokens` is 131072 and can go up to 1048576, but reasoning tokens inflate response size — monitor your usage and adjust upward if needed. K3.1 Pro uses `reasoning_effort` (`low` / `high` / `max`, default `max`), not the K2.x `thinking` parameter; add `"reasoning_effort": "low"` to `requestBody` if answers take too long. The proxy forwards the value untouched.
87
- >
88
- > Both models are in the snippet above — delete the entry you don't plan to use. K3 accepts the same `requestBody`; it only supports `reasoning_effort: "max"`.
89
73
 
90
74
  ### 2. API key
91
75
 
@@ -129,30 +113,29 @@ Set in `.env` at the repo root (the proxy `import 'dotenv/config'` automatically
129
113
  - Forwards your `Authorization` header upstream.
130
114
  - Rewrites plain-chat requests to `temperature: 1`, `top_p: 0.95`.
131
115
  - For **K2.5 / K2.6**: rewrites tool-enabled requests to `thinking: { type: "disabled" }`, `temperature: 0.6`, `top_p: 0.95`.
132
- - For **K3 / K3.1 Pro**: keeps thinking enabled (both are always-thinking and use `reasoning_effort`, not `thinking`); rewrites to `temperature: 1`, `top_p: 0.95`. Does **not** inject a `thinking` block — the K3 family rejects it. Every `kimi-k3*` slug is handled this way, including `kimi-k3.1-pro` and `kimi-k3.1-pro-highspeed`.
116
+ - For **K3.1 Pro**: keeps thinking enabled (always-thinking, uses `reasoning_effort`, not `thinking`); rewrites to `temperature: 1`, `top_p: 0.95`. Does **not** inject a `thinking` block — K3.1 Pro rejects it. Every `kimi-k3*` slug is handled this way, including `kimi-k3.1-pro` and `kimi-k3.1-pro-highspeed`.
133
117
  - Preserves streaming responses (SSE).
134
118
  - Writes redacted request summaries to `debug_log/kimi-proxy.ndjson`.
135
119
 
136
120
  ## Notes
137
121
 
138
- - **Kimi K3.1 Pro** (released October 23, 2026) is the current flagship: a post-trained K3 with substantially larger post-training compute, 3T class, 1M context, open weights promised by November 4, 2026. Always-thinking — chain-of-thought **cannot** be disabled; tune depth with `reasoning_effort` (`low` / `high` / `max`, default `max`). Do **not** send the K2.x `thinking` parameter. `tool_choice` supports `auto`, `required`, and named tools. `max_completion_tokens` defaults to 131072 and can go up to 1048576. Sampling is fixed at `temperature=1`, `top_p=0.95`, `n=1`, `presence_penalty=0`, `frequency_penalty=0` — the proxy's enforced `1` / `0.95` are the only accepted values. Vision supports text + image plus video via `ms://<file-id>` URLs; public image URLs are rejected, so pass base64 or an uploaded file id. See the [K3.1 Pro quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart).
139
- - **Kimi K3** (released July 16, 2026) remains served but is superseded by K3.1 Pro: 2.8T params, 1M context, always-thinking with `reasoning_effort` `max` only. It shares K3.1 Pro's request shape, so switching between them needs only a new `id` — no proxy or `requestBody` change. See the [K3 quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-quickstart).
122
+ - **Kimi K3.1 Pro** (released October 23, 2026) is the current flagship: 3T class, 1M context, with substantially larger post-training compute, open weights promised by November 4, 2026. Always-thinking — chain-of-thought **cannot** be disabled; tune depth with `reasoning_effort` (`low` / `high` / `max`, default `max`). Do **not** send the K2.x `thinking` parameter. `tool_choice` supports `auto`, `required`, and named tools. `max_completion_tokens` defaults to 131072 and can go up to 1048576. Sampling is fixed at `temperature=1`, `top_p=0.95`, `n=1`, `presence_penalty=0`, `frequency_penalty=0` — the proxy's enforced `1` / `0.95` are the only accepted values. Vision supports text + image plus video via `ms://<file-id>` URLs; public image URLs are rejected, so pass base64 or an uploaded file id. See the [K3.1 Pro quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart).
140
123
  - **Kimi Platform rebranding:** the Kimi Platform (formerly Moonshot) rebranded to `platform.kimi.ai` in 2026 — `api.moonshot.ai` still resolves and the proxy targets it unchanged.
141
- - **`tool_choice`:** K3 and K3.1 Pro support `auto`, `required`, and named tools. The proxy preserves the value and only enforces temperature/top_p.
142
- - **Do not send `thinking`:** the K3 family rejects the legacy K2.x `thinking` parameter — use `reasoning_effort` at the top level of the request body instead.
124
+ - **`tool_choice`:** K3.1 Pro supports `auto`, `required`, and named tools. The proxy preserves the value and only enforces temperature/top_p.
125
+ - **Do not send `thinking`:** K3.1 Pro rejects the legacy K2.x `thinking` parameter — use `reasoning_effort` at the top level of the request body instead.
143
126
 
144
127
  ## Troubleshooting
145
128
 
146
- | Symptom | Likely cause | Fix |
147
- | ------------------------------------------------------ | ---------------------------- | ------------------------------------------------- |
148
- | "Connection refused" on chat | Proxy not running | `npm run proxy:kimi` |
149
- | `invalid temperature: only 1 is allowed` | Direct path without proxy | Use the proxy |
150
- | `invalid top_p: only 0.95 is allowed` | Direct path without proxy | Use the proxy |
151
- | `thinking is enabled but reasoning_content is missing` | Tool turn with thinking on | Verify `KIMI_PROXY_DISABLE_THINKING_WITH_TOOLS=1` |
152
- | `unsupported parameter: thinking` with K3 | K3 doesn't accept `thinking` | Use `reasoning_effort` instead; proxy detects K3 |
153
- | Reasoning can't be turned off | K3.1 Pro always thinks | Set `reasoning_effort` to `low` instead |
154
- | Model not in VS Code picker | Config not reloaded | Restart VS Code |
155
- | `tool_choice=required` rejected (K2.x) | Model limitation | Use `auto` only for K2.x; K3 supports `required` |
129
+ | Symptom | Likely cause | Fix |
130
+ | ------------------------------------------------------ | ---------------------------------- | ------------------------------------------------------- |
131
+ | "Connection refused" on chat | Proxy not running | `npm run proxy:kimi` |
132
+ | `invalid temperature: only 1 is allowed` | Direct path without proxy | Use the proxy |
133
+ | `invalid top_p: only 0.95 is allowed` | Direct path without proxy | Use the proxy |
134
+ | `thinking is enabled but reasoning_content is missing` | Tool turn with thinking on | Verify `KIMI_PROXY_DISABLE_THINKING_WITH_TOOLS=1` |
135
+ | `unsupported parameter: thinking` with K3.1 Pro | K3.1 Pro doesn't accept `thinking` | Use `reasoning_effort` instead; proxy detects the model |
136
+ | Reasoning can't be turned off | K3.1 Pro always thinks | Set `reasoning_effort` to `low` instead |
137
+ | Model not in VS Code picker | Config not reloaded | Restart VS Code |
138
+ | `tool_choice=required` rejected (K2.x) | Model limitation | Use `auto` only for K2.x; K3.1 Pro supports `required` |
156
139
 
157
140
  ## Pricing
158
141
 
@@ -161,6 +144,5 @@ For the cross-provider comparison, see [docs/pricing.md](../pricing.md). Moonsho
161
144
  | Model | Input (cache miss) | Cached input | Output |
162
145
  | --------------- | ------------------ | ------------ | ----------- |
163
146
  | `kimi-k3.1-pro` | $2.00 / 1M | $0.20 / 1M | $10.00 / 1M |
164
- | `kimi-k3` | $3.00 / 1M | $0.30 / 1M | $15.00 / 1M |
165
147
 
166
- > K3 and K3.1 Pro pricing is flat (no tiering by context length), and caching is automatic for repeated prefixes — a 90% discount on cached input. Cache writes are billed separately by TTL (5 minutes / 1 hour) on both models: K3's write rates are $3.00/M (5 min) and $6.00/M (1 h). The `kimi-k3.1-pro-highspeed` SKU costs more than K3.1 Pro. Check the [official pricing table](https://platform.kimi.ai/docs/pricing/chat) for current write rates and the Highspeed rate.
148
+ > K3.1 Pro pricing is flat (no tiering by context length), and caching is automatic for repeated prefixes — a 90% discount on cached input. Cache writes are billed separately by TTL (5 minutes / 1 hour). The `kimi-k3.1-pro-highspeed` SKU costs more than K3.1 Pro. Check the [official pricing table](https://platform.kimi.ai/docs/pricing/chat) for current write rates and the Highspeed rate.
@@ -1,44 +1,43 @@
1
1
  # MiniMax — VS Code Custom Endpoint Setup Guide
2
2
 
3
- > **TL;DR:** **MiniMax-M3.1-Flash** (released October 19, 2026) is the current M-series Flash model and needs no proxy. Use `thinking: { "type": "adaptive" }` + `reasoning_split: true` + `reasoning_effort` in `requestBody` so the model can reason and the response arrives in a clean OpenAI format (`reasoning_content`, kept separate from `content`). **M3.1 Flash cannot disable thinking** — `thinking: { "type": "disabled" }` and `reasoning_effort: "none"` both return HTTP 400 (`requires adaptive thinking`, code 2013) — so tune depth with `reasoning_effort` (`low` / `medium` / `high` / `xhigh` / `max`, default `max`). `MiniMax-M3` uses the same request shape, still accepts `thinking: { "type": "disabled" }` as a soft hint.
3
+ > **TL;DR:** **MiniMax-M3.1-Flash** (released October 19, 2026) is the current M-series Flash model and needs no proxy. Use `thinking: { "type": "adaptive" }` + `reasoning_split: true` + `reasoning_effort` in `requestBody` so the model can reason and the response arrives in a clean OpenAI format (`reasoning_content`, kept separate from `content`). **M3.1 Flash cannot disable thinking** — `thinking: { "type": "disabled" }` and `reasoning_effort: "none"` both return HTTP 400 (`requires adaptive thinking`, code 2013) — so tune depth with `reasoning_effort` (`low` / `medium` / `high` / `xhigh` / `max`, default `max`).
4
4
  >
5
5
  > The same `url`, `model id`, and `requestBody` work for **both** Pay-as-You-Go (account-balance billing) and Token Plan (monthly/annual subscription) — only the API key in the secret field changes.
6
6
  >
7
7
  > **🆕 Try a MiniMax Copilot extension first** (no JSON editing):
8
8
  >
9
- > - **Token Plan subscription** (`sk-cp-…`): the [**klarkxy/minimax-vscode**](https://marketplace.visualstudio.com/items?itemName=klarkxy.minimax-vscode-copilot) extension is the most feature-rich — reasoning visibility, usage dashboard, status-bar quota monitoring, one-click region switching, and M3 1M context toggling. Subscription keys required.
9
+ > - **Token Plan subscription** (`sk-cp-…`): the [**klarkxy/minimax-vscode**](https://marketplace.visualstudio.com/items?itemName=klarkxy.minimax-vscode-copilot) extension is the most feature-rich — reasoning visibility, usage dashboard, status-bar quota monitoring, and one-click region switching. Subscription keys required.
10
10
  > - **Pay-as-You-Go** (Open Platform API Key): the [**tugudush/minimax-copilot**](https://github.com/tugudush/minimax-copilot) extension is a PAYG-native alternative — same Anthropic-compatible API surface, but designed for pay-as-you-go keys. Collapsible reasoning block, region auto-detection, walkthrough for first-time setup. See [docs/research/minimax-vscode-extension.md](../research/minimax-vscode-extension.md) for the full side-by-side comparison.
11
11
  >
12
12
  > If you'd rather not install an extension, the manual setup below works with both PAYG and Token Plan keys via the OpenAI-compatible endpoint.
13
13
 
14
14
  ## At a Glance
15
15
 
16
- | Field | Value |
17
- | ------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------- |
18
- | Mode | **Direct** (no proxy) |
19
- | Billing | **Pay-as-You-Go** _or_ **Token Plan subscription** (same config) |
20
- | Vision | ✅ Yes (image + video) |
21
- | Tool calling | ✅ Yes |
22
- | Context | 1M (M3 guarantees 512K) |
23
- | Max output | 131072 recommended (524288 max) |
24
- | Required `requestBody` | M3.1 Flash: `thinking: { type: "adaptive" }, reasoning_split: true, reasoning_effort: "high"` · M3: `thinking: { type: "adaptive" }, reasoning_split: true` |
25
- | Endpoint (international) | `https://api.minimax.io/v1/chat/completions` |
26
- | Endpoint (China) | `https://api.minimaxi.com/v1/chat/completions` |
27
- | Auth | `Authorization: Bearer $MINIMAX_API_KEY` (or Subscription Key, see [Setup](#setup)) |
16
+ | Field | Value |
17
+ | ------------------------ | --------------------------------------------------------------------------------------------- |
18
+ | Mode | **Direct** (no proxy) |
19
+ | Billing | **Pay-as-You-Go** _or_ **Token Plan subscription** (same config) |
20
+ | Vision | ✅ Yes (image + video) |
21
+ | Tool calling | ✅ Yes |
22
+ | Context | 1M |
23
+ | Max output | 131072 recommended (524288 max) |
24
+ | Required `requestBody` | M3.1 Flash: `thinking: { type: "adaptive" }, reasoning_split: true, reasoning_effort: "high"` |
25
+ | Endpoint (international) | `https://api.minimax.io/v1/chat/completions` |
26
+ | Endpoint (China) | `https://api.minimaxi.com/v1/chat/completions` |
27
+ | Auth | `Authorization: Bearer $MINIMAX_API_KEY` (or Subscription Key, see [Setup](#setup)) |
28
28
 
29
29
  ### Models
30
30
 
31
31
  | Model | Vision | Context | Max output | Notes |
32
32
  | -------------------- | ------ | ------- | ------------------------------- | ------------------------------------------------------------------------------------------------------ |
33
33
  | `MiniMax-M3.1-Flash` | ✅ Yes | 1M | 131072 recommended (524288 max) | Announced October 19, 2026; tunable thinking depth via `reasoning_effort`; open weights promised later |
34
- | `MiniMax-M3` | ✅ Yes | 1M | 131072 | Multimodal frontier coding model; adaptive reasoning; PAYG and Token Plan share config |
35
34
 
36
35
  ## Quick Start
37
36
 
38
37
  1. **Edit `chatLanguageModels.json`** — add the MiniMax block below.
39
38
  2. **Set your `MINIMAX_API_KEY`** via Command Palette → **Chat: Manage Language Models**.
40
39
  3. **Configure the Utility Small Model** — Open Settings → search **"Chat: Utility Small Model"** → pick your fastest model (e.g., DeepSeek V4.1 Flash or MiMo V2.6 Flash). [Why?](../../README.md#4-configure-the-utility-small-model)
41
- 4. **Restart VS Code** and pick "MiniMax M3.1 Flash (vision)" (or "MiniMax M3") in the chat picker.
40
+ 4. **Restart VS Code** and pick "MiniMax M3.1 Flash (vision)" in the chat picker.
42
41
 
43
42
  ## Setup
44
43
 
@@ -82,32 +81,16 @@
82
81
  "temperature": 1,
83
82
  "top_p": 0.95
84
83
  }
85
- },
86
- {
87
- "id": "MiniMax-M3",
88
- "name": "MiniMax M3 (vision)",
89
- "url": "https://api.minimax.io/v1/chat/completions",
90
- "toolCalling": true,
91
- "vision": true,
92
- "streaming": true,
93
- "maxInputTokens": 1048576,
94
- "maxOutputTokens": 131072,
95
- "requestBody": {
96
- "thinking": { "type": "adaptive" },
97
- "reasoning_split": true,
98
- "temperature": 1,
99
- "top_p": 0.95
100
- }
101
84
  }
102
85
  ]
103
86
  }
104
87
  ```
105
88
 
106
- > **Model IDs are case-sensitive.** Use exactly `MiniMax-M3.1-Flash` and `MiniMax-M3` (capital `M`s, lowercase `i`). Both entries can live in the same provider group — delete whichever one you don't need. Do **not** invent suffixed ids or duplicate an existing one: VS Code forwards each `id` as the upstream `model` parameter, and the API only knows the two ids above.
89
+ > **Model IDs are case-sensitive.** Use exactly `MiniMax-M3.1-Flash` (capital `M`s, lowercase `i`). Do **not** invent suffixed ids: VS Code forwards each `id` as the upstream `model` parameter, and the API only knows that id.
107
90
 
108
91
  ### 2. Select your billing mode
109
92
 
110
- MiniMax exposes the **same** `https://api.minimax.io/v1/chat/completions` endpoint and the **same** model ids (`MiniMax-M3.1-Flash` / `MiniMax-M3`) to both billing modes. The only field that changes is the API key — the rest of `chatLanguageModels.json` stays identical.
93
+ MiniMax exposes the **same** `https://api.minimax.io/v1/chat/completions` endpoint and the **same** model id (`MiniMax-M3.1-Flash`) to both billing modes. The only field that changes is the API key — the rest of `chatLanguageModels.json` stays identical.
111
94
 
112
95
  | Billing mode | Key name in console | Where to get it | How usage is metered |
113
96
  | ------------- | --------------------------------------- | ----------------------------------------------------------------- | ----------------------------------------------- |
@@ -151,31 +134,31 @@ M3.1 Flash always thinks — `thinking: { "type": "disabled" }` and `reasoning_e
151
134
  }
152
135
  ```
153
136
 
154
- `reasoning_split: true` keeps thinking out of `content` and on `reasoning_content`. Setting it to `false` is not supported on M3.1 Flash, and the server already returns thinking separately by default. `temperature` and `top_p` are accepted but ignored on the M3 family (fixed at `1` and `0.95`); they are kept here only so every provider block in this repo has the same shape.
137
+ `reasoning_split: true` keeps thinking out of `content` and on `reasoning_content`. Setting it to `false` is not supported on M3.1 Flash, and the server already returns thinking separately by default. `temperature` and `top_p` are accepted but ignored on M3.1 Flash (fixed at `1` and `0.95`); they are kept here only so every provider block in this repo has the same shape.
155
138
 
156
139
  ## Notes
157
140
 
158
- - **`thinking: adaptive` + `reasoning_split: true`** is the recommended pair on both models: the server returns thinking on the `reasoning_content` field, which keeps `content` clean for VS Code. On M3.1 Flash the `thinking` field is effectively fixed at `adaptive` (the server rejects `disabled`); on M3, `disabled` is a soft hint — the model may still reason internally and emit `<think>` tags / `reasoning_content`, so the setting changes the response layout more than the model's actual behavior.
159
- - **`reasoning_effort` only does something on M3.1 Flash.** M3 ignores it. Values are `low` / `medium` / `high` / `xhigh` / `max`, default `max`; `none` is rejected.
160
- - **`temperature: 1`, `top_p: 0.95`** are accepted but **ignored** on both models (the M3 family is fixed at those values); `top_p` must still be inside `(0, 1]` if you send it.
141
+ - **`thinking: adaptive` + `reasoning_split: true`** is the recommended pair: the server returns thinking on the `reasoning_content` field, which keeps `content` clean for VS Code. The `thinking` field is effectively fixed at `adaptive` (the server rejects `disabled`).
142
+ - **`reasoning_effort` controls thinking depth.** Values are `low` / `medium` / `high` / `xhigh` / `max`, default `max`; `none` is rejected.
143
+ - **`temperature: 1`, `top_p: 0.95`** are accepted but **ignored** (M3.1 Flash is fixed at those values); `top_p` must still be inside `(0, 1]` if you send it.
161
144
  - **Interleaved thinking and tool loops:** MiniMax recommends sending the complete assistant message — including `reasoning_content` — back in history so the reasoning chain stays continuous. VS Code does not preserve `reasoning_content` across tool turns, so expect shallower continuity on long agent runs. This is a quality trade-off, not an error: unlike Kimi K2.7, M3.1 Flash does not reject a tool result whose assistant message lacks reasoning content.
162
- - **Max output:** `max_completion_tokens` is recommended at 131072 and accepts up to 524288 on both models; thinking tokens count toward that limit.
145
+ - **Max output:** `max_completion_tokens` is recommended at 131072 and accepts up to 524288; thinking tokens count toward that limit.
163
146
  - **Rate limits:** 200 RPM / 10M TPM. Input tokens above 512K are available in limited quantity for a limited time.
164
147
 
165
148
  ## Troubleshooting
166
149
 
167
150
  | Symptom | Likely cause | Fix |
168
151
  | -------------------------------------------- | ----------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
169
- | Model not in picker | Config not reloaded, or wrong casing | Restart VS Code; verify the model ID is exactly `MiniMax-M3.1-Flash` or `MiniMax-M3` |
152
+ | Model not in picker | Config not reloaded, or wrong casing | Restart VS Code; verify the model ID is exactly `MiniMax-M3.1-Flash` |
170
153
  | Reasoning leaks into `content` | Missing `reasoning_split` | Add `reasoning_split: true` to `requestBody` |
171
154
  | 400 `requires adaptive thinking` (2013) | `thinking: { "type": "disabled" }` or `reasoning_effort: "none"` sent to M3.1 Flash | Remove the override; lower `reasoning_effort` instead. On M3.1 Flash thinking cannot be disabled. |
172
155
  | 400 `reasoning_split` cannot be false | `reasoning_split: false` sent to M3.1 Flash | Use `reasoning_split: true` (or omit it); split output is mandatory on this model. |
173
- | 400 / 2013 with `service_tier` | `service_tier` sent to M3.1 Flash | Remove `service_tier` — only `MiniMax-M3` accepts that field |
156
+ | 400 / 2013 with `service_tier` | `service_tier` sent to M3.1 Flash | Remove `service_tier` — M3.1 Flash does not accept that field |
174
157
  | 401 Unauthorized | API key region mismatch | Use the endpoint that matches your key's region |
175
158
  | 401 `token is unusable` / 1004 | Wrong key type on the endpoint — PAYG key against a Token Plan quota or vice versa | The two keys are not interchangeable. Switch the secret in **Chat: Manage Language Models** to the correct key for the billing mode you intend to use. |
176
159
  | 402 / 429 after a burst of agent turns | Token Plan 5-hour or weekly quota exhausted; not a PAYG balance issue | Wait for the quota window to reset, buy a Credits top-up at `platform.minimax.io/user-center/payment/credits`, upgrade plan, or swap the secret to a PAYG key to continue. |
177
160
  | 429 rate-limited | Concurrent sessions exceeded 200 RPM / 10M TPM | Reduce concurrent agent sessions |
178
- | Vision request returns 400 | Vision is not supported on the legacy M2.x line | Use `MiniMax-M3.1-Flash` or `MiniMax-M3` |
161
+ | Vision request returns 400 | Vision is not supported on the legacy M2.x line | Use `MiniMax-M3.1-Flash` |
179
162
  | Token Plan key works in curl but not VS Code | VS Code is still resolving the old `${input:chat.lm.secret.<id>}` reference | Re-open **Chat: Manage Language Models**, re-paste the Subscription Key, then restart VS Code |
180
163
  | China account gets 401 on `api.minimax.io` | Key region is China but the URL points to international (or vice versa) | Switch the `url` to `https://api.minimaxi.com/v1/chat/completions` for China, and get the Subscription Key from `platform.minimaxi.com` |
181
164
 
@@ -187,19 +170,6 @@ For the cross-provider comparison, see [docs/pricing.md](../pricing.md). MiniMax
187
170
  | ------------ | ----------------- | ------------------ | ---------- |
188
171
  | All requests | $0.015 / 1M | $0.15 / 1M | $0.60 / 1M |
189
172
 
190
- The `MiniMax-M3` rates below are **effective** rates — a permanent 50% off discount halves the list prices:
191
-
192
- ### Standard tier
193
-
194
- | Token range | Input (Cache Hit) | Input (Cache Miss) | Output |
195
- | --------------------- | ----------------- | ------------------ | ---------- |
196
- | ≤ 512K input tokens | $0.06 / 1M | $0.30 / 1M | $1.20 / 1M |
197
- | > 512K input tokens\* | $0.12 / 1M | $0.60 / 1M | $2.40 / 1M |
198
-
199
- \* Input tokens above 512K are available in limited quantity for a limited time.
200
-
201
- > **Permanent 50% off:** A standing 50% discount applies to all MiniMax-M3 pay-as-you-go usage (verified June 9, 2026). The effective rates above are post-discount. List prices (pre-discount) are 2× higher: $0.12 / $0.60 / $2.40 per 1M at ≤512K input.
202
-
203
173
  ### Token Plan (subscription)
204
174
 
205
175
  The Token Plan is a **monthly or annual** subscription with a shared multimodal usage bar (text + image + audio + video + music share one quota).
@@ -1,6 +1,6 @@
1
1
  # Qwen (DashScope) — VS Code Custom Endpoint Setup Guide
2
2
 
3
- > **TL;DR:** `qwen3.8-max` (vision) works both direct and via the local proxy. OpenRouter's live rankings card shows unversioned Qwen3.8 Max at **#3 with 53.4**, while the exact `qwen/qwen3.8-max-0902` API metadata row is **45.4**; the public API no longer exposes the unversioned ID. The proxy forwards the upstream `model` value unchanged, so use the exact model ID supported by the endpoint you choose. The proxy gives you dynamic thinking suppression: reasoning stays ON in plain chat but turns OFF automatically when tools are invoked. The direct path is simpler if you don't need reasoning in chat. Use the **OpenAI-compatible** endpoint on your **workspace-dedicated** domain ([details](#which-endpoint-to-use)).
3
+ > **TL;DR:** `qwen3.8-max` (vision) works both direct and via the local proxy. The proxy forwards the upstream `model` value unchanged, so use the exact model ID supported by the endpoint you choose. The proxy gives you dynamic thinking suppression: reasoning stays ON in plain chat but turns OFF automatically when tools are invoked. The direct path is simpler if you don't need reasoning in chat. Use the **OpenAI-compatible** endpoint on your **workspace-dedicated** domain ([details](#which-endpoint-to-use)).
4
4
 
5
5
  ## At a Glance
6
6
 
@@ -27,93 +27,6 @@
27
27
 
28
28
  > The live `chatLanguageModels.json` points Qwen at the local proxy by default; the direct DashScope URL is shown below for users who prefer a static `enable_thinking: false` setup.
29
29
 
30
- ### OpenRouter snapshot records
31
-
32
- These are separate OpenRouter catalog entries, not alternative labels for one model:
33
-
34
- | OpenRouter slug | Snapshot | AA Intelligence Index | Notes |
35
- | ------------------------ | -------- | --------------------- | ----------------------------------------------------------------------------- |
36
- | `qwen/qwen3.8-max` | 0803 | **53.4 (#3)** | Current ranking-card result; exact unversioned API entry is no longer exposed |
37
- | `qwen/qwen3.8-max-0902` | 0902 | **45.4** | September snapshot; current exact-ID API metadata |
38
- | `qwen/qwen3.8-max-prime` | 0923 | — | Higher-throughput variant at 2× the 0902 rates; no AA composite yet |
39
-
40
- The DashScope custom-endpoint snippets below use the provider model ID `qwen3.8-max`. Do not replace it with the OpenRouter `qwen/qwen3.8-max-0902` slug unless you also change the upstream URL to OpenRouter and have verified that path independently.
41
-
42
- OpenRouter's `qwen/qwen3.8-max-prime` SKU (0923) is a higher-throughput variant billed at 2× the 0902 list rates (**$4 / $0.50 cache read / $12** per 1M tokens). No AA composite is published for it yet, and it is not part of the validated DashScope setup.
43
-
44
- ### Optional OpenRouter snapshot proxy
45
-
46
- The proxy forwards the `model` field without mapping it, so it can also be pointed at OpenRouter for these catalog entries. This is an **optional, unvalidated OpenRouter path**; it is separate from the validated DashScope setup above and requires an OpenRouter API key.
47
-
48
- 1. Point the proxy at OpenRouter — pick **one** of these launch methods:
49
- - **From the repo (`.env`):** add this to the repo-root `.env` file, then start the proxy with `npm run proxy:qwen`:
50
-
51
- ```dotenv
52
- QWEN_UPSTREAM_URL=https://openrouter.ai/api/v1/chat/completions
53
- ```
54
-
55
- > `npm run proxy:qwen` always runs with cwd = repo root, so the proxy's `import 'dotenv/config'` picks up this `.env` automatically.
56
-
57
- - **Via npx:** `npx copilot-custom-endpoint qwen` forks the same proxy, but loads `.env` from **whatever directory you run it in**. If that directory has no `.env` with `QWEN_UPSTREAM_URL`, the proxy silently falls back to the legacy DashScope upstream and OpenRouter slugs like `qwen/qwen3.8-max` will fail against it. Either run it from a directory containing a suitable `.env`, or set the variable in the shell first (see next option). Note that `npx` runs the **published** package version, not your local working tree.
58
-
59
- - **Inline env var (Git Bash / bash):** shell variables take precedence over `.env` (dotenv does not override variables already in `process.env`), so this works from any directory without editing `.env`:
60
-
61
- ```bash
62
- QWEN_UPSTREAM_URL=https://openrouter.ai/api/v1/chat/completions npx copilot-custom-endpoint qwen
63
- ```
64
-
65
- PowerShell equivalent:
66
-
67
- ```powershell
68
- $env:QWEN_UPSTREAM_URL = 'https://openrouter.ai/api/v1/chat/completions'
69
- npx copilot-custom-endpoint qwen
70
- ```
71
-
72
- 2. Add a separate provider group using the OpenRouter key and these model entries:
73
-
74
- ```json
75
- {
76
- "name": "Qwen (OpenRouter snapshots)",
77
- "vendor": "customendpoint",
78
- "apiKey": "",
79
- "apiType": "chat-completions",
80
- "models": [
81
- {
82
- "id": "qwen/qwen3.8-max",
83
- "name": "Qwen 3.8 Max (0803)",
84
- "url": "http://127.0.0.1:3458/v1/chat/completions",
85
- "toolCalling": true,
86
- "vision": true,
87
- "streaming": true,
88
- "maxInputTokens": 991000,
89
- "maxOutputTokens": 131072
90
- },
91
- {
92
- "id": "qwen/qwen3.8-max-0902",
93
- "name": "Qwen 3.8 Max (0902)",
94
- "url": "http://127.0.0.1:3458/v1/chat/completions",
95
- "toolCalling": true,
96
- "vision": true,
97
- "streaming": true,
98
- "maxInputTokens": 991000,
99
- "maxOutputTokens": 131072
100
- },
101
- {
102
- "id": "qwen/qwen3.8-max-prime",
103
- "name": "Qwen 3.8 Max Prime (0923)",
104
- "url": "http://127.0.0.1:3458/v1/chat/completions",
105
- "toolCalling": true,
106
- "vision": true,
107
- "streaming": true,
108
- "maxInputTokens": 991000,
109
- "maxOutputTokens": 131072
110
- }
111
- ]
112
- }
113
- ```
114
-
115
- Set the API key through **Chat: Manage Language Models**, then verify the proxy with `curl http://127.0.0.1:3458/healthz`. Do not run this route at the same time as the default DashScope route unless each process uses a different `QWEN_PROXY_PORT` and log path.
116
-
117
30
  ## Quick Start
118
31
 
119
32
  ### With Proxy (recommended)
@@ -280,7 +193,7 @@ The proxy detects active tool use by examining the conversation state (a `"tool"
280
193
  ## Notes
281
194
 
282
195
  - **Vision (`qwen3.8-max`)** uses OpenAI-compatible `content` array format. Base64 data URIs work reliably; external image URLs may fail if DashScope can't reach them. If a drag-and-drop image fails to load, providing the absolute file path (e.g. `c:\path\to\image.png`) in the prompt is a reliable workaround.
283
- - **Qwen 3.8 reasoning:** `qwen3.8-max` enables reasoning by default and supports `reasoning_effort` values `low`, `medium`, and `xhigh` (default). The direct snippet disables reasoning for stable VS Code tool loops; the proxy leaves it on for plain chat and suppresses it when tool activity is detected. The OpenRouter ranking card and 0902 API metadata are separate source records; the benchmark difference does not mean the DashScope config should silently switch IDs.
196
+ - **Qwen 3.8 reasoning:** `qwen3.8-max` enables reasoning by default and supports `reasoning_effort` values `low`, `medium`, and `xhigh` (default). The direct snippet disables reasoning for stable VS Code tool loops; the proxy leaves it on for plain chat and suppresses it when tool activity is detected.
284
197
  - **Thinking trade-off:** Direct = thinking always off (loops stable, no reasoning visible). Proxy = thinking on in plain chat, off in tool turns.
285
198
  - **`tool_choice` only supports `auto`** — don't override it (VS Code's default is `auto`).
286
199
 
package/docs/pricing.md CHANGED
@@ -20,7 +20,7 @@ All prices below are in **USD per 1M tokens** (non-cached). To convert to AI cre
20
20
 
21
21
  All models are listed together below, sorted by Cost per intelligence ascending (lower is better). Models without a Cost per intelligence are ordered by estimated session cost ascending. Session cost assumes ~10K input + ~2K output tokens per turn, 50 turns.
22
22
 
23
- **All AA Intelligence Index values in this table were re-based from v4.3.2 to v5.0 on October 31, 2026.** AA’s v5.0 composite aggregates a different evaluation set than v4.3.2, so scores are not comparable with earlier snapshots of this file. Nine rows — GPT-5.6 Luna, Qwen 3.8 Max (0803), Grok 4.6, GPT-6 Sol, Claude Sonnet 5, GPT-5.6 Terra, GPT-5.6 Sol, Claude Opus 5, and Claude Fable 5 — have no v5.0 composite on AA’s current leaderboard; their Cost per intelligence is `—` and they are ordered after the scored rows by estimated session cost.
23
+ **All AA Intelligence Index values in this table were re-based from v4.3.2 to v5.0 on October 31, 2026.** AA’s v5.0 composite aggregates a different evaluation set than v4.3.2, so scores are not comparable with earlier snapshots of this file.
24
24
 
25
25
  | Model | Provider | Cost per intelligence | Intelligence Score | Input (per 1M) | Cached input | Output (per 1M) | Est. session | Context window |
26
26
  | ---------------------------- | --------- | --------------------- | ------------------ | ----------------------------- | ----------------------------- | ----------------------------- | ------------ | -------------- |
@@ -31,7 +31,6 @@ All models are listed together below, sorted by Cost per intelligence ascending
31
31
  | **MiniMax M3.1 Flash** | MiniMax | **~$0.0047** | **28.9** | $0.15 | $0.015 | $0.60 | ~$0.14 | 1M |
32
32
  | **DeepSeek V4.1 Flash 1017** | DeepSeek | **~$0.0084** | **32** | $0.30 | $0.006 | $1.20 | ~$0.27 | 1M |
33
33
  | **MiMo V2.6 Pro** | Xiaomi | **~$0.0102** | **29.9** | $0.435 | $0.0036 | $0.87 | ~$0.30 | 1M |
34
- | **MiniMax M3** | MiniMax | **~$0.0144** | **18.7** | $0.60 (≤512K) / $1.20 (>512K) | $0.12 (≤512K) / $0.24 (>512K) | $2.40 (≤512K) / $4.80 (>512K) | ~$0.27 | 1M |
35
34
  | **GPT-6.1 Terra** | OpenAI | **~$0.0196** | **51** | $1.00 | $0.05 | $5.00 | ~$1.00 | 1M |
36
35
  | **Gemini 3.8 Flash** | Google | **~$0.0246** | **30.5** | $0.75 | $0.075 | $3.75 | ~$0.75 | 1M |
37
36
  | **Muse Spark 1.3** | Meta | **~$0.0288** | **36.4** | $1.25 | $0.15 | $4.25 | ~$1.05 | 1M |
@@ -40,50 +39,36 @@ All models are listed together below, sorted by Cost per intelligence ascending
40
39
  | **Claude Sonnet 5.5** | Anthropic | **~$0.0369** | **54.2** | $2.00 | $0.10 | $10.00 | ~$2.00 | 1M |
41
40
  | **Gemini 4 Argon** | Google | **~$0.0397** | **50.4** | $2.00 | $0.10 | $10.00 | ~$2.00 | 1M |
42
41
  | **Kimi K3.1 Pro** | Moonshot | **~$0.0435** | **46** | $2.00 | $0.20 | $10.00 | ~$2.00 | 1M |
42
+ | **GPT-6 Sol** | OpenAI | **~$0.0455** | **44** | $2.00 | $0.20 | $10.00 | ~$2.00 | 1M |
43
43
  | **Grok 4.7** | xAI | **~$0.0457** | **35** | $2.00 | $0.50 | $6.00 | ~$1.60 | 500K |
44
- | **Qwen 3.8 Max (0902)** | DashScope | **~$0.0476** | **33.6** | $2.00 | $0.25 | $6.00 | ~$1.60 | 1M |
44
+ | **Qwen 3.8 Max** | DashScope | **~$0.0476** | **33.6** | $2.00 | $0.25 | $6.00 | ~$1.60 | 1M |
45
45
  | **Claude Opus 5.5** | Anthropic | **~$0.0685** | **58.4** | $4.00 | $0.20 | $20.00 | ~$4.00 | 1M |
46
46
  | **MiMo V2.6 Pro UltraSpeed** | Xiaomi | **~$0.1018** | **29.9** | $4.35 | $0.036 | $8.70 | ~$3.05 | 1M |
47
- | **Kimi K3** | Moonshot | **~$0.1724** | **17.4** | $3.00 | $0.30 | $15.00 | ~$3.00 | 1M |
47
+ | **Claude Fable 5.2** | Anthropic | **~$0.1231** | **65** | $8.00 | $0.20 | $40.00 | ~$8.00 | 1M |
48
48
  | **GPT-6 Astra** | OpenAI | **~$0.1757** | **56.9** | $10.00 | $1.00 | $50.00 | ~$10.00 | 1M |
49
49
  | **Claude Fable 5.1** | Anthropic | **~$0.2012** | **49.7** | $10.00 | $0.25 | $50.00 | ~$10.00 | 1M |
50
- | **GPT-5.6 Luna** | OpenAI | — | — | $0.20 | $0.02 | $1.20 | ~$0.22 | 1M |
51
- | **Qwen 3.8 Max (0803)** | DashScope | — | — | $2.00 | $0.25 | $6.00 | ~$1.60 | 1M |
52
- | **Grok 4.6** | xAI | — | — | $2.00 | $0.50 | $6.00 | ~$1.60 | 500K |
53
- | **GPT-6 Sol** | OpenAI | — | — | $2.00 | $0.20 | $10.00 | ~$2.00 | 1M |
54
- | **Claude Sonnet 5** | Anthropic | — | — | $2.00 | $0.20 | $10.00 | ~$2.00 | 1M |
55
- | **GPT-5.6 Terra** | OpenAI | — | — | $2.00 | $0.20 | $12.00 | ~$2.20 | 1M |
56
- | **GPT-5.6 Sol** | OpenAI | — | — | $4.00 | $0.40 | $20.00 | ~$4.00 | 1M |
57
- | **Claude Opus 5** | Anthropic | — | — | $5.00 | $0.50 | $25.00 | ~$5.00 | 1M |
58
- | **Claude Fable 5** | Anthropic | — | — | $10.00 | $1.00 | $50.00 | ~$10.00 | 1M |
59
50
 
60
51
  Cost per intelligence = estimated session cost ÷ Intelligence Index score. Session cost assumes ~10K input + ~2K output tokens per turn, 50 turns.
61
52
 
62
53
  > **Notes:**
63
54
  >
64
- > - **DeepSeek** input pricing shown is the **peak cache miss** price. Peak cache hits are $0.006/M for V4.1 Flash; off-peak rates are half. The October 17, 2026 **V4.1-Flash-1017** build keeps those rates unchanged, is served under the same `deepseek-flash` ID, and supports both thinking (default) and non-thinking modes. AA measures it at 216 tokens/s and very verbose (210M output tokens to run the Intelligence Index versus a 140M median), so real sessions can cost more than this table's modeled token counts. See the [DeepSeek pricing page](https://api-docs.deepseek.com/quick_start/pricing) for the full schedule.
65
- > - **MiMo** input pricing shown is the **cache miss** price. Cache hits are essentially free for V2.6 Pro ($0.0036/M, ~120× cheaper), V2.6 Flash ($0.0028/M, ~50× cheaper), and V2.6 Pro UltraSpeed ($0.036/M, ~120× cheaper). A Xiaomi price cut took effect on 2026-05-27.
66
- > - **Anthropic (Claude)** cache reads and cache writes are billed separately. Claude Opus 5.5 cache reads are $0.20/M and Sonnet 5.5 cache reads are $0.10/M (both 0.05× base input); cache writes are $5/M for 5 minutes or $8/M for 1 hour on Opus 5.5 and $2.50/M or $4/M on Sonnet 5.5. Fast mode costs $8/M input and $40/M output. Opus 5.5, Opus 5, Sonnet 5 / 5.5, Haiku 5.5, and Fable 5 / 5.1 use a newer tokenizer that produces approximately 30% more tokens for the same text. **Claude Haiku 5.5 is the only current Claude model with prompt-length tiering**: prompts up to 100,000 tokens cost $0.10 / $0.01 cache read / $0.50 per 1M, prompts over 100,000 tokens cost $0.50 / $0.05 / $2.50, cache writes are $0.125/M (5 min) and $0.20/M (1 h) below the threshold, and batch pricing is $0.05 / $0.25 (≤100K) and $0.25 / $1.25 (>100K); the table above shows the sub-100K rates that Anthropic says cover ~90% of Haiku-class requests. Anthropic states Haiku 5.5 costs about 75% less on average to run than Haiku 4.5. Claude Sonnet 5.5 keeps Sonnet 5's $2 / $10 rates, and Anthropic made them permanent: the Sonnet 5 increase to $3 / $15 scheduled for September 1 was cancelled — and as of October 7, 2026 its cache reads were halved to $0.10/M (batch $1 / $5). Artificial Analysis measured Sonnet 5.5 at 138.7 tokens/s but very verbose (410M output tokens to run the Intelligence Index vs an 88M median), and Haiku 5.5 at 244 tokens/s but also very verbose (440M vs a 100M median), so real sessions can cost more than this table's modeled token counts; Anthropic reports up to 30% fewer tokens per task than Sonnet 5.
67
- > - **OpenAI** models support cached input at 0.1× base input rate — except **GPT-6.1 Sol**, whose cache hits are 0.05× ($0.10/M, half of GPT-6 Sol's $0.20/M). GPT-6 Sol, GPT-6.1 Sol, and Luna also bill cache writes separately at 1.25× input. The GPT-6 and GPT-6.1 rows show standard rates through 272K input; for requests >272K, input and cache-read rates double and output is 1.5×. **GPT-6.1 Terra** (October 2026) is the newest GPT-6.1 SKU and is priced flat at **$1.00 / $0.05 cache read (95% discount) / $5.00** per 1M tokens — half of GPT-6.1 Sol’s input and output rates. Cache writes are $1.25/M at 30-minute retention or $2.50/M at 12-hour retention, requests above 272K input tokens are billed at 2× input and cache rates with 1.5× output for the whole request, and Batch/Flex pricing is 50% lower.
68
- > - **Google (Gemini)** — **Gemini 4 Argon** is priced at Google's introductory rates: $2 / $0.10 implicit cache / $10 per 1M tokens through November 30, 2026, then $4 / $0.20 / $20. Output prices include thinking tokens. Only implicit caching is supported — a flat 95% discount on cached input for prompts of at least 4,096 tokens (explicit caching is not available). Gemini 4 Argon is not offered on the free tier or the Google AI Pro plan, and GitHub Copilot bills it at provider list pricing (Pro+, Max, Business, and Enterprise; not Pro).
69
- > - **Qwen** models use **tiered pricing** — determined by total input tokens per request. Prices above are for non-thinking mode.
70
- > - **Kimi** official tables list **Cache Hit before Cache Miss** (opposite order to our table; Kimi K3 and K3.1 Pro also bill cache writes separately by TTL — K3 writes are $3.00/M for a 5-minute TTL and $6.00/M for 1 hour). The rows transpose them so "Input" = cache miss and "Cached input" = cache hit. **Kimi K3.1 Pro** (October 23, 2026) is **$2.00 / $0.20 cached / $10.00** per 1M with flat rates and automatic prefix caching (90% cache discount). It is always-thinking: `reasoning_effort` accepts `low` / `high` / `max` (default `max`) and chain-of-thought cannot be disabled. Artificial Analysis measures it at ~81 tokens/s and very verbose (150M output tokens to run the Intelligence Index versus an 87M median), at $2.11 per index task, so real sessions can cost more than the modeled ~$2.00. See [docs/models/kimi.md](models/kimi.md). **Kimi K3** (July 2026) is the previous flagship — superseded by K3.1 Pro but still served at $3.00 / $0.30 cached / $15.00.
71
- > - **DashScope** offers a **free quota** of 1M input + 1M output tokens per model, valid for 90 days.
72
- > - **MiniMax M3** uses **tiered pricing** — input price doubles above 512K input tokens. Cache hits are priced at 20% of the input rate ($0.12/M ≤512K, $0.24/M >512K). A **permanent 50% off** discount applies to all MiniMax-M3 pay-as-you-go usage, making the effective rates half the list prices above.
73
- > - **MiniMax M3.1 Flash** is priced **flat** at $0.15 / $0.015 cache read / $0.60 per 1M tokens — one rate for all prompt lengths, a 90% cache discount, and **no `service_tier` support** (the API rejects that field for this model). Thinking cannot be disabled: `thinking: { "type": "disabled" }` or `reasoning_effort: "none"` returns HTTP 400, and depth is tuned with `reasoning_effort` (`low` / `medium` / `high` / `xhigh` / `max`, default `max`). Artificial Analysis measures it at 96 tokens/s and very verbose (180M output tokens to run the Intelligence Index vs a 100M median), so real sessions can cost more than the modeled ~$0.14. See [docs/models/minimax.md](models/minimax.md).
74
- > - **GLM** models support prompt caching — cache hits are priced at $0.26/M for 5.3 and $0.03/M for 5.3 Flash (list; $0.015/M during the launch promo).
75
- > - **MiMo** offers a **Token Plan** subscription model with discounted rates and a free cache-writing promotion. MiMo V2.6 Flash's **26.7** composite comes from OpenRouter's Artificial Analysis v5.0 metadata (AA also measures it at 55.4 tokens/s and very verbose); Xiaomi's published launch benchmarks for it are recorded in [docs/benchmarks.md](benchmarks.md).
76
- > - **Meta (Muse Spark)** — the table shows the **Standard** tier of the Meta Model API (`muse-spark-1.3`): $1.25 / $0.15 cached input / $4.25 per 1M tokens, with automatic prefix caching and no long-context surcharge. Meta's day-one **Contributor** SKU (`muse-spark-1.3-contributor`) is $0.10 / $0.002 / $0.20 per 1M — about 12× cheaper on input and 21× on output — but Meta may use those prompts and completions to train future models, so it is excluded from this comparison the way promotions are. Integrated web-search grounding is billed separately at $2.50 per 1,000 queries on both tiers, and rate limits are 3,000 RPM / 4M TPM (Standard) versus 100 RPM / 3M TPM (Contributor). Muse Spark 1.3 is closed and API-only (Meta has announced future open weights without a date), and its audio understanding is degraded versus Muse Spark 1.2 — Meta recommends 1.2 or Muse Voice Transcribe for audio.
55
+ > - **DeepSeek:** the table shows peak cache-miss pricing; off-peak rates are half. See the [DeepSeek pricing page](https://api-docs.deepseek.com/quick_start/pricing).
56
+ > - **MiMo:** the table shows cache-miss pricing; cache hits are far cheaper (see the Cached input column). A Token Plan subscription with discounted rates is also available.
57
+ > - **Anthropic:** cache reads and writes are billed separately. Haiku 5.5 is the only Claude model with prompt-length tiering; the table shows its ≤100K rates. Fast mode is $8 / $40 per 1M.
58
+ > - **OpenAI:** GPT-6 Sol, GPT-6.1 Sol, and Luna bill cache writes at 1.25× input. Requests over 272K input cost 2× input and cache rates, with 1.5× output.
59
+ > - **Gemini 4 Argon:** the $2 / $10 introductory rates run through November 30, 2026, then rise to $4 / $20. Output includes thinking tokens.
60
+ > - **Kimi K3.1 Pro and MiniMax M3.1 Flash:** both always think, so control cost with `reasoning_effort`. See [docs/models/kimi.md](models/kimi.md) and [docs/models/minimax.md](models/minimax.md).
61
+ > - **GLM:** cache hits are $0.26/M for 5.3 and $0.03/M for 5.3 Flash at list.
62
+ > - **Qwen:** pricing is tiered by input tokens per request; the table shows non-thinking rates. DashScope offers 1M input + 1M output free tokens per model for 90 days.
63
+ > - Artificial Analysis measures several models as very verbose, so real sessions can cost more than the modeled token counts.
77
64
  > - For typical Copilot chat usage (short-to-medium prompts), you'll almost always fall in the lowest pricing tier.
78
65
 
79
66
  > **How long does 7,000 credits last?** A Pro+ subscriber running 50-turn sessions can estimate monthly capacity by multiplying each modeled session cost by 100 to convert it to AI credits, then mixing models as needed.
80
67
 
81
- > Prices last verified: October 31, 2026. Always check the official pages for the latest rates:
68
+ > Prices last verified: November 10, 2026. Always check the official pages for the latest rates:
82
69
  >
83
70
  > - [GitHub Copilot models & pricing](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing)
84
71
  > - [OpenAI pricing](https://openai.com/api/pricing/)
85
- > - [OpenAI GPT-5.6 announcement](https://openai.com/index/gpt-5-6/)
86
- > - [OpenAI GPT-5.6 price update](https://openai.com/index/advancing-the-price-performance-frontier-with-gpt-5-6/)
87
72
  > - [OpenAI GPT-6 Sol and Luna announcement](https://openai.com/index/introducing-gpt-6-sol-and-luna/)
88
73
  > - [OpenAI GPT-6 Sol API details](https://developers.openai.com/api/docs/models/gpt-6-sol)
89
74
  > - [OpenAI GPT-6 Luna API details](https://developers.openai.com/api/docs/models/gpt-6-luna)
@@ -93,6 +78,9 @@ Cost per intelligence = estimated session cost ÷ Intelligence Index score. Sess
93
78
  > - [OpenAI GPT-6.1 Terra API details](https://developers.openai.com/api/docs/models/gpt-6.1-terra)
94
79
  > - [GitHub Copilot GPT-6.1 Terra release](https://github.blog/changelog/2026-10-26-gpt-6-1-terra-in-github-copilot)
95
80
  > - [Anthropic (Claude) pricing](https://platform.claude.com/docs/en/about-claude/pricing)
81
+ > - [Anthropic Claude Fable 5.2 announcement](https://www.anthropic.com/claude-fable-and-mythos-5-2)
82
+ > - [GitHub Copilot Claude Fable 5.2 release](https://github.blog/changelog/2026-11-09-claude-fable-5-2-in-github-copilot)
83
+ > - [Artificial Analysis Claude Fable 5.2](https://artificialanalysis.ai/models/claude-fable-5-2)
96
84
  > - [Anthropic Claude Opus 5.5 announcement](https://www.anthropic.com/claude-opus-5-5)
97
85
  > - [Anthropic Opus 5.5 API docs](https://platform.claude.com/docs/en/models/opus-5-5/overview)
98
86
  > - [Anthropic Claude Sonnet 5.5 announcement](https://www.anthropic.com/claude-sonnet-5-5)
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "copilot-custom-endpoint",
3
- "version": "3.6.0",
3
+ "version": "3.7.1",
4
4
  "description": "Local proxies for VS Code Copilot custom endpoints — Kimi K2, Qwen 3.x & MiMo V2.6",
5
5
  "license": "MIT",
6
6
  "type": "module",