npm - free-coding-models - Versions diffs - 0.5.98 → 0.5.99 - Mend

free-coding-models 0.5.98 → 0.5.99

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.

Files changed (6) hide show

package/README.md +9 -9
package/changelog/v0.5.99.md +17 -0
package/package.json +1 -1
package/sources.js +105 -68
package/web/dist/assets/{index-CP-gd42H.js → index-Zr5Q6fX9.js} +3 -3
package/web/dist/index.html +1 -1

package/README.md CHANGED Viewed

@@ -6,7 +6,7 @@
 <p align="center">
   <strong>Find the fastest free coding model in seconds.</strong><br>
-  Live latency, stability and verdicts for 253 models from 25 free AI providers, then install the one you pick straight into your favorite coding tool.<br><br>
+  Live latency, stability and verdicts for 252 models from 24 free AI providers, then install the one you pick straight into your favorite coding tool.<br><br>
   <strong>Works with:</strong> OpenCode CLI / Desktop / WebUI, OpenClaw, Crush, Goose, Aider, Kilo CLI, Qwen Code, OpenHands, Amp, Hermes, Continue, Cline, Xcode, Pi, ZCode, ForgeCode, Copilot, jcode, Caveman Code and more.
 </p>
@@ -24,7 +24,7 @@ free-coding-models
 ```
 <p align="center">
-  <sub>Then create a free account on one of the <a href="#-providers">25 providers</a> to grab an API key.</sub>
+  <sub>Then create a free account on one of the <a href="#-providers">24 providers</a> to grab an API key.</sub>
 </p>
 <p align="center">
@@ -35,7 +35,7 @@ free-coding-models
 ## 💡 Why this tool?
-There is a large catalog of free and free-limited coding models (**25 providers / 253 live models**, generated from [`sources.js`](./sources.js)). Which one is fastest *right now*? Which one is actually stable, versus just lucky on the last ping?
+There is a large catalog of free and free-limited coding models (**24 providers / 252 live models**, generated from [`sources.js`](./sources.js)). Which one is fastest *right now*? Which one is actually stable, versus just lucky on the last ping?
 `free-coding-models` (FCM) answers that by pinging every model in parallel, showing live latency, and computing a **live Stability Score (0-100)** combining p95 latency, jitter, spike rate and uptime. Average latency alone is misleading: a model that randomly spikes to 6 seconds is not reliable.
@@ -87,18 +87,18 @@ free-coding-models --fiable              # print the single most reliable model
 ## 🟢 Providers
-**25 active providers / 253 live models**, sorted by live model count. Top 8:
+**24 active providers / 252 live models**, sorted by live model count. Top 8:
 | Provider | Models | Best tier | Env var |
 |----------|--------|-----------|---------|
-| [Pollinations AI](https://enter.pollinations.ai) | 24 | S+ | `POLLINATIONS_API_KEY` |
+| [Pollinations AI](https://enter.pollinations.ai) | 25 | S+ | `POLLINATIONS_API_KEY` |
 | [Alibaba DashScope](https://modelstudio.console.alibabacloud.com) | 22 | S+ | `DASHSCOPE_API_KEY` |
-| [onomeo](https://onomeo.com/dashboard) | 19 | S+ | `ONOMEO_API_KEY` |
-| [Ollama Cloud](https://ollama.com/settings/keys) | 18 | S+ | `OLLAMA_API_KEY` |
+| [Ollama Cloud](https://ollama.com/settings/keys) | 19 | S+ | `OLLAMA_API_KEY` |
+| [Kilo](https://kilo.ai) | 16 | S+ | `KILO_API_KEY` |
 | [OpenRouter](https://openrouter.ai/keys) | 16 | S+ | `OPENROUTER_API_KEY` |
-| [Kilo](https://kilo.ai) | 15 | S+ | `KILO_API_KEY` |
+| [onomeo](https://onomeo.com/dashboard) | 15 | S+ | `ONOMEO_API_KEY` |
+| [NVIDIA NIM](https://build.nvidia.com) | 14 | S+ | `NVIDIA_API_KEY` |
 | [OVHcloud AI](https://endpoints.ai.cloud.ovh.net) | 14 | S+ | `OVH_AI_ENDPOINTS_ACCESS_TOKEN` |
-| [NVIDIA NIM](https://build.nvidia.com) | 13 | S+ | `NVIDIA_API_KEY` |
 > 🧾 **What "free" means here:** free is a property of the *(provider, model)* pair, never of the provider as a whole. A row is listed only when that exact model id costs $0 to call through that provider (permanent free tier, `:free` variant, or free plan), verified live at audit time. The same open-weights model can be free on one host and paid on another - paid siblings are deliberately excluded. Full breakdown and badge legend: [`docs/providers.md`](./docs/providers.md).

package/changelog/v0.5.99.md ADDED Viewed

@@ -0,0 +1,17 @@
+# Changelog v0.5.99 - 2026-10-17
+### Changed
+- 📡 **Full 24-provider audit** (weekly): 253 → 252 models. 29 models removed, 28 added, 18 config fixes, all verified against live provider APIs and official docs on 2026-10-17. `audit_state.json` refreshed for all live providers (the stale `github-models` entry was dropped along with `zai`).
+- 🚀 **Reflection Beam 1.0 everywhere**: the new 501B MoE agentic coder (SWE-bench Verified 80.9-81.6%) landed free on 4 providers the same week - Kilo, OpenCode Zen, Vercel AI Gateway and Ollama Cloud. It is now the top free coder in the catalog.
+- ➕ **New free models**: NVIDIA NIM adds Nemotron 3.5 Super VL (71.07% SWE, multimodal) and StepFun Step 5 Preview (600B MoE, 1M ctx); XingChen Xing4.0 29B (75.0% SWE) joins OpenRouter and Kilo; MiniMax M3 (80.5% SWE, 1M ctx) on SambaNova was proposed but NOT re-added - a `/v1/models` listing is not free-tier evidence and the official Free Tier table still excludes it (the 2026-10-10 verdict stands, re-check next audit). Scaleway adds GLM-5.3, DeepSeek V4.1 Flash and brings back Qwen3 Coder 30B. Pollinations adds GPT-5.4 (1M), Grok 4.6, Grok 4.3, Qwen3.8 2.4T A95B, Mistral Large 3 and Nemotron 3 Ultra. LLM7 gets MiniMax M2.7 back. Vercel AI Gateway adds Nemotron 3.5 Lightning Free and the stealth cat-jam slot. onomeo swaps space-bunny-alpha for the new stealth glyph-cluster.
+- ➖ **Dead or paid-flipped models removed**: NVIDIA killed Kimi K3 (deprecated 2026-10-14, gone from /v1/models). Novita's free tier collapsed from 8 to 2 models (the Ling free promo and GLM/Qwen/MiniMax free tiers all ended; Apodex 1.1 Mini moved in from OpenRouter as its replacement). Pollinations moved 5 models behind paid Pollen (qwen-coder, glm-5.2, grok-4.7, step-5-preview, nova-2-lite). OpenCode Zen rotated out 4 dead promo slots (step-5-preview-free, ling-3.1/3.0-flash-free, muse-spark-1.3-contributor). Kilo lost step-5-preview-free and ling-3.1-flash (both now paid). Vercel delisted ling-3.1-flash-free. onomeo delisted 5 models (gemini-3.8-flash, glm-5.3-flash, glm-4.7-flash, glm-4.6v-flash, space-bunny-alpha). Mistral retired Leanstral 1.5.1.
+- 🗑️ **Z.ai provider retired from the catalog**: the GLM Coding Plan free window ended 2026-10-07 and the plan is now paid-only (Lite $18/mo+), so the provider has zero free models left. The `zai` export stays as an empty array for backwards compat (same pattern as GitHub Models) and ZAI_API_KEY support remains in the code for existing configs. Free GLM-5.3-Flash lives on via NVIDIA NIM, OpenRouter, Pollinations, OrcaRouter and onomeo. 24 active providers now.
+- 🔧 **Context windows and ids fixed against live APIs**: 18 corrections, including the Requesty silent rename `novita/inclusionai/ling-3.1-flash` → `novita/ling-3.1-flash`, Ollama glm-5.2 976k → 1M and nemotron-3-nano 1M → 256k, LLM7 kimi-k2.6 262k → 256k, Mistral ministral-3b 131k → 128k, plus 7 Requesty and 3 Pollinations ctx normalizations.
+- ⏳ **OVHcloud Mistral Small 3.2 gets an auto-expiry date** (2026-11-04): OVH announced its decommission on 2026-08-04; the model still answers today so it stays listed, and the new `deprecatedAfter` tuple element removes it from the catalog automatically after the shutdown date.
+- 📊 **Quota labels refreshed**: Cerebras is now honest about its free tier ("$5 credits / 30 days, card required" instead of "generous dev tier"), onomeo's rate limit dropped to 30 req/5h, Novita shows "2 zero-price models left".
+- 🧪 **Tests updated for the catalog rotation** (same pattern as previous audits): router/installer fixtures moved from the dead `moonshotai/kimi-k3` NVIDIA slot to `z-ai/glm-5.3`, and the daemon-level zai normalizer subtest was folded into the existing unit coverage since the provider left the free catalog. 1185 tests pass.
+### Fixed
+- Nothing else; this release is a pure catalog audit.

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "free-coding-models",
-  "version": "0.5.98",
+  "version": "0.5.99",
   "description": "Find the fastest coding LLM models in seconds — ping free models from multiple providers, pick the best one for OpenCode, Cursor, or any AI coding assistant.",
   "keywords": [
     "nvidia",

package/sources.js CHANGED Viewed

@@ -49,9 +49,11 @@ export const nvidiaNim = [
   // Removed (2026-09-05): moonshotai/kimi-k2.6 (Kimi K2.6) - Model page returns 404 and model is absent from the NVIDIA model catalog; could not verify existence
   // Removed (2026-08-30): deepseek-ai/deepseek-v4-pro (DeepSeek V4 Pro) — 410 Gone per NVIDIA NIM forum; replaced by deepseek-v4-flash:0731 (forums.developer.nvidia.com/t/deepseek-v4-pro-flash-removed/379558)
   // Removed (2026-09-21): deepseek-ai/deepseek-v4-flash-0731 (DeepSeek V4 Flash) — NVIDIA deprecation banner on the model page: deprecated 2026-09-19, no longer supported after 2026-09-21; DeepSeek retired V4 Flash in favor of V4.1 Flash (not registered on NIM)
-  ['moonshotai/kimi-k3', 'Kimi K3', 'S+', '76.8%', '1M'], // Fixed (2026-09-21): tier 'S' → 'S+' + sweScore '-' → '76.8%' (tracker-sourced SWE-bench Verified; 76.8% is S+ on the documented scale)
+  // Removed (2026-10-17 audit): moonshotai/kimi-k3 (Kimi K3) — officially deprecated on the NIM model card (deprecated 2026-10-14, unsupported after 2026-10-15) and absent from /v1/models; no Moonshot model remains usable on NIM (kimi-k2.6 endpoint was shut down 2026-07-07 and its catalog page 404s)
   ['z-ai/glm-5.3-flash', 'GLM-5.3-Flash', 'S+', '-', '1M'], // Added (2026-09-22) — live on NIM /v1/models (needed a 230s cold start on first probe)
   ['z-ai/glm-5.3', 'GLM 5.3', 'S+', '-', '1M'], // Fixed (2026-09-29 audit): ctx '200k' → '1M' (NVIDIA docs state 1,048,576 tokens)
+  ['nvidia/nemotron-3.5-super-vl-120b-a12b', 'Nemotron 3.5 Super VL', 'S+', '71.07%', '262k'], // Added (2026-10-17 audit) — new hosted model, live on /v1/models + build.nvidia.com; multimodal 120B/12B MoE, SWE-bench Verified 71.07% per the NVIDIA model card
+  ['stepfun-ai/step-5-preview', 'Step 5 Preview', 'S', '-', '1M'], // Added (2026-10-17 audit) — new StepFun flagship preview (600B/27B-active MoE), live on /v1/models, 1M ctx; mirrors the Step 5 rollout on other gateways
   // Removed (2026-09-28 audit re-check): deepseek-ai/deepseek-v4-pro, deepseek-ai/deepseek-v4-flash-0731, minimaxai/minimax-m2.7, stepfun-ai/step-3.7-flash, nvidia/nemotron-3-nano-30b-a3b re-appear in the public catalog but were all probed 410/404 here 2026-09-05..21 — ghost catalog entries again, still excluded (issue #181 policy)
   // Removed (2026-08-30): stepfun-ai/step-3.7-flash (Step 3.7 Flash) — 410 Gone per NVIDIA NIM TUI ping (no replacement listed; superseded by step-3.7-flash via Routeway `step-3.7-flash:free`)
   ['nvidia/nemotron-3-ultra-550b-a55b', 'Nemotron 3 Ultra', 'S+', '71.9%', '1M'],
@@ -168,16 +170,18 @@ export const openrouter = [
   ['nvidia/nemotron-3-ultra-550b-a55b:free', 'Nemotron 3 Ultra', 'S+', '71.9%', '1M'],
   ['poolside/laguna-xs-2.1:free', 'Poolside Laguna XS 2.1', 'S+', '70.9%', '262k'],
   ['poolside/laguna-s-2.1:free', 'Poolside Laguna S 2.1', 'S+', '-', '262k'],
+  ['xingchen-agi/xing4.0-29b-a4b:free', 'Xing4.0 29B A4B', 'S+', '75.0%', '262k'], // Added (2026-10-17 audit) — new :free entry (TeleAI Xing4.0 agentic coder, SWE-bench Verified 75.0% per the published model card)
   // Removed (2026-09-15): minimax/minimax-m2.7:free (MiniMax M2.7) — no longer free on OpenRouter
   // Removed (2026-09-15): minimax/minimax-m3:free (MiniMax M3) — no longer free on OpenRouter
   // Removed (2026-09-28 audit): z-ai/glm-5.2:free (GLM-5.2) — gone from the live :free catalog (now paid-only)
   // ── S tier — SWE-bench Verified 60–70% ──
   ['cohere/north-mini-code:free', 'North Mini Code', 'S', '-', '256k'],
   ['nvidia/nemotron-3-super-120b-a12b:free', 'Nemotron 3 Super', 'S', '60.5%', '262k'],
-  ['inclusionai/ling-3.1-flash', 'Ling 3.1 Flash', 'S', '-', '262k'], // Added (2026-10-10 audit) — new $0/$0 in the live catalog (560B-A25B MoE); replaces the paid-only ling-3.0-flash-sante
+  // Removed (2026-10-17 audit): inclusionai/ling-3.1-flash (Ling 3.1 Flash) — no longer $0/$0: now bills $0.075/M in + $0.225/M out with no :free variant; reverses the 2026-10-10 add
   // Removed (2026-10-10 audit): qwen/qwen3.8-27b:free (Qwen3.8 27B) — deleted from the live :free catalog (Kilo mirror dropped it the same day); no free Qwen replacement
   // ── A+ tier — SWE-bench Verified 50–60% ──
   ['nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free', 'Nemotron 3 Omni', 'A+', '52.0%', '256k'],
+  ['nvidia/nemotron-3.5-super-vl:free', 'Nemotron 3.5 Super VL', 'S+', '71.1%', '1M'], // Added (2026-10-17 audit) — new :free entry, 1M ctx per the live API; SWE score follows the same model on NVIDIA NIM (71.07% rounded per family reads), S+ on the documented scale
   ['google/gemma-4-31b-it:free', 'Gemma 4 31B', 'A+', '52.0%', '262k'],
   ['google/gemma-4-26b-a4b-it:free', 'Gemma 4 26B MoE', 'A', '38.0%', '262k'],
   // Removed (2026-09-28 audit): nex-agi/nex-n2.5-pro:free — nex-agi dropped from the live :free catalog entirely
@@ -188,7 +192,7 @@ export const openrouter = [
   ['thinkingmachines/inkling:free', 'Inkling', 'B+', '-', '1M'], // Added (2026-09-02)
   // Removed (2026-10-10 audit): inclusionai/ling-3.0-flash-sante:free — :free variant removed, base id now bills per-token; replacement: inclusionai/ling-3.1-flash ($0/$0)
   // Removed (2026-10-10 audit): stealth/space-bunny-alpha — stealth preview rotated out of the live list, no stealth model currently live (next stealth slot: glyph-cluster on Kilo/Vercel)
-  ['apodex/apodex-1.1-mini:free', 'Apodex 1.1 Mini', 'B+', '-', '262k'], // Added (2026-10-10 audit) — new in the live :free catalog, 262k ctx
+  // Removed (2026-10-17 audit): apodex/apodex-1.1-mini:free — delisted from OpenRouter entirely (no free or paid listing in the live API); the model moved to Novita as a zero-price partner launch
   ['openrouter/free', 'Free Models Router', 'A', '-', '200k'], // Added (2026-09-29 audit) — OpenRouter's meta-router, $0 but routes to a RANDOM free model per request (utility pick, quality varies)
   // Removed (2026-09-28 audit): nex-agi/nex-n2.5-mini:free — nex-agi dropped from the live :free catalog entirely
   // Removed (2026-09-28 audit): inclusionai/ling-3.0-flash-vl:free — :free variant removed from the live catalog (base model now billed)
@@ -235,9 +239,8 @@ export const mistral = [
   ['ministral-14b-2512', 'Ministral 3 14B', 'B+', '-', '256k'], // Fixed (2026-09-16): ministral-3-14b-25-12 → ministral-14b-2512 (live /v1/models, ctx 262144)
   // ── B tier — SWE-bench Verified 20–30% ──
   ['ministral-8b-2512', 'Ministral 3 8B', 'B', '-', '256k'], // Fixed (2026-09-16): ministral-3-8b-25-12 → ministral-8b-2512 (live /v1/models, ctx 262144)
-  ['labs-leanstral-1-5-1', 'Leanstral 1.5.1', 'B', '-', '256k'], // Added (2026-10-10 audit) — successor of leanstral-1-5, live on /v1/models (free-tier key probe); Lean 4 theorem proving, niche coding use
-  // Removed (2026-10-10 audit): labs-leanstral-1-5 (Leanstral 1.5) — docs mark it deprecated 2026-09-29 with retirement 2026-09-30 (passed); Labs-gated on the free plan. Replacement: labs-leanstral-1-5-1
-  ['ministral-3b-2512', 'Ministral 3 3B', 'B', '-', '131k'], // Fixed (2026-09-16): ministral-3-3b-25-12 → ministral-3b-2512; ctx 256k → 131k (max_context_length 131072; 2026-09-28 audit re-confirmed 131k)
+  // Removed (2026-10-17 audit): labs-leanstral-1-5-1 (Leanstral 1.5.1) — Labs Leanstral line deprecated 2026-09-29 and retired 2026-09-30 per the docs.mistral.ai deprecation table; free-tier probe now returns 403 tier_not_allowed (the signature the repo documented for paid-only models)
+  ['ministral-3b-2512', 'Ministral 3 3B', 'B', '-', '128k'], // Fixed (2026-10-17 audit): ctx '131k' → '128k' (official max_context_length 128000 per docs.mistral.ai); id fixed 2026-09-16
   // Removed (2026-09-16): mistral-small-creative-25-12 (Mistral Small Creative) — absent from the live catalog
   // ── Coding models (codestral-2508 aliases, live-verified 2026-09-22 via api.mistral.ai/v1/models) ──
   ['mistral-code-latest', 'Mistral Code', 'A', '-', '256k'], // Added (2026-09-22): aliases codestral-2508 / codestral-latest; ctx 256000
@@ -260,7 +263,9 @@ export const codestral = [
 export const scaleway = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '256k'], // Fixed (2026-08-13): ctx '1M' → '256k' (Serverless tier per official catalog)
+  ['glm-5.3', 'GLM 5.3', 'S+', '-', '256k'], // Added (2026-10-17 audit) — free Serverless on the official supported-models page; score pending a published SWE-bench Verified figure
   ['deepseek-v4-flash-0731', 'DeepSeek V4 Flash', 'S+', '-', '256k'], // Added (2026-08-13)
+  ['deepseek-v4.1-flash', 'DeepSeek V4.1 Flash', 'S+', '-', '256k'], // Added (2026-10-17 audit) — free Serverless on the official supported-models page (upstream retired V4-Flash-0731 on 2026-09-10 but Scaleway still serves both)
   // Removed (2026-09-05): devstral-2-123b-instruct-2512 (Devstral 2 123B) - Deprecated 2026-07-01, End of Life 2026-08-01; after EOL the model is no longer accessible on Generative APIs Serverless
   // ── S tier — SWE-bench Verified 60–70% ──
   ['qwen3.5-397b-a17b', 'Qwen3.5 400B VLM', 'S+', '76.2%', '250k'],
@@ -271,6 +276,7 @@ export const scaleway = [
   // Removed (2026-09-15): qwen3-coder-30b-a3b-instruct (Qwen3 Coder 30B) — Deprecated 2026-07-01, EOL 2026-10-01; replacement: qwen3.6-35b-a3b
   // Audit (2026-09-28): all 11 entries confirmed live; qwen3-coder-30b-a3b-instruct still EOL 2026-10-01, not re-added
   ['qwen3.6-35b-a3b', 'Qwen3.6 35B MoE', 'S+', '73.4%', '256k'],
+  ['qwen3-coder-30b-a3b-instruct', 'Qwen3 Coder 30B', 'A+', '51.6%', '128k'], // Re-added (2026-10-17 audit) — live free Serverless again on the official supported-models page (the 2026-10-01 EOL verdict is reversed upstream); 51.6% is the official Qwen SWE-bench Verified for the 30B coder
   // Removed (2026-09-05): holo2-30b-a3b (Holo2 30B) - Deprecated 2026-07-09, End of Life 2026-08-09; after EOL the model is no longer accessible on Generative APIs Serverless
   ['gemma-4-26b-a4b-it', 'Gemma 4 26B MoE', 'A+', '-', '256k'],
   // Removed (2026-09-02): gemma-4-31b-it (Gemma 4 31B IT) — Dedicated tier only, not available on Serverless
@@ -308,12 +314,16 @@ export const googleai = [
 ]
 // 📖 ZAI source - https://open.z.ai
-// 📖 Free tier is limited to Flash models; paid GLM models are intentionally excluded.
-// 📖 Verified live (2026-08-23) via ping test: glm-4.5-flash and glm-4.6v-flash still serve free;
-// 📖 glm-4.7-flash is free but was returning "overloaded" 429s; API /models lists only 9 text models.
+// 📖 ⚠️ NO FREE TIER LEFT (2026-10-17 audit) — the GLM Coding Plan devpack free window
+// 📖 (zero-quota 23:00-09:00 SGT, ran 2026-09-03 → 2026-10-07) is over and the plan is
+// 📖 now paid-only (Lite/Pro/Max, from $18/mo; docs.z.ai/devpack/overview). GLM-5.3-Flash
+// 📖 bills $0.15/$0.50 per 1M on the pay-per-token API. Free GLM-5.3-Flash remains on
+// 📖 NVIDIA NIM / OpenRouter / Pollinations / OrcaRouter / onomeo gateways.
+// 📖 Kept as an empty array so downstream provider-metadata/config code that imports
+// 📖 `zai` and references the 'zai' key (id normalizers, ZAI_API_KEY) doesn't crash; the
+// 📖 entry is also commented out of the `sources` map below so it won't appear in the catalog.
 export const zai = [
-  // ── S+ tier — SWE-bench Verified ≥70% ──
-  ['zai/glm-5.3-flash', 'GLM-5.3-Flash', 'S+', '-', '1M'], // Added (2026-09-02) — free via Coding Plan devpack campaign (runs to 2026-10-07); zero quota 23:00-09:00 SGT
+  // Removed (2026-10-17 audit): zai/glm-5.3-flash (GLM-5.3-Flash) — the Coding Plan devpack free window ended 2026-10-07 and the GLM Coding Plan is paid-only (Lite $18/mo+); the id now serves a paid model. Free GLM-5.3-Flash: z-ai/glm-5.3-flash on NVIDIA NIM
   // Removed (2026-09-29 audit): zai/glm-5.3 (GLM-5.3) — paid-only on the API ($1.4/$4.4 per 1M) and needs the paid $18/mo Coding Plan; docs pricing page has no free column for it (provider premise: Flash models only)
   // Removed (2026-09-21): zai/glm-5.2 + zai/glm-5.1 — Coding Plan requests for both are now silently redirected to GLM-5.3 (official docs.z.ai plan update), so the ids no longer serve a distinct free model; both remain paid-API models
   // Removed (2026-09-28 audit): zai/glm-5 — legacy id, Coding Plan auto-routes GLM-5 requests to GLM-5.3
@@ -432,7 +442,7 @@ export const ovhcloud = [
   // Removed (2026-07-27): Mistral-7B-Instruct-v0.3 (Mistral 7B Instruct) — no longer in catalog
   // Removed (2026-08-13): Mistral-Nemo-Instruct-2407 (Mistral Nemo) — no longer in OVHcloud public catalog
   // Removed (2026-09-21): Mistral-Small-3.2-24B-Instruct-2506 + Mistral-Nemo-Instruct-2407 + Mistral-7B-Instruct-v0.3 re-removed — none of the three Mistral models appear on the official AI Endpoints catalog page; the 2026-09-21 morning re-adds were erroneous
-  ['Mistral-Small-3.2-24B-Instruct-2506',       'Mistral Small 3.2 24B','B+', '20.0%', '131k'], // Re-added (2026-09-29 audit) — served by the live router API (marketing page hides it); score 20.0% reused from the same model id on Scaleway
+  ['Mistral-Small-3.2-24B-Instruct-2506',       'Mistral Small 3.2 24B','B+', '20.0%', '131k', null, '2026-11-04'], // ⚠️ DEPRECATED — OVHcloud decommission 2026-11-04 (announced 2026-08-04, API 404s after that date); 7th tuple element auto-filters it from the catalog. Re-added (2026-09-29 audit) — served by the live router API (marketing page hides it); score 20.0% reused from the same model id on Scaleway
   ['Qwen3.5-9B',                                'Qwen3.5 9B',           'B+', '30.0%', '262k'],
   ['Qwen2.5-VL-72B-Instruct', 'Qwen2.5-VL 72B', 'S', '-', '32k'], // Added (2026-08-13)
   // ── Embeddings ──
@@ -457,18 +467,23 @@ export const opencodeZen = [
   // Removed (2026-09-29 audit): deepseek-v4-flash-free (DeepSeek V4 Flash Free) — live chat probe returns 400 "Upstream request failed: Model is unavailable" and it is gone from the docs free table; the 2026-09-21 re-add said "re-verify at next audit", verified dead
   // Removed (2026-10-10 audit): mimo-v2.5-free (MiMo-V2.5 Free) — live chat probe returns 401 ModelError "not supported" and it is absent from the /v1/models list (docs table from Oct 8 is stale on this one); replacement: mimo-v2.6-flash-free
   ['mimo-v2.6-flash-free', 'MiMo-V2.6 Flash Free', 'S+', '-', '200k'], // Added (2026-09-22) — new free promo model on the live Zen /v1/models list
-  ['step-5-preview-free', 'Step 5 Preview Free', 'A+', '-', '200k'], // Added (2026-10-10 audit) — new in the docs free table / live Zen list (matches step-5-preview rollout on Kilo/llm7/pollinations)
-  ['ling-3.1-flash-free', 'Ling 3.1 Flash Free', 'B+', '-', '262k'], // Added (2026-10-10 audit) — new free id in the docs free table; replaces ling-3.0-flash-* slots rotating out across gateways
+  // Removed (2026-10-17 audit): step-5-preview-free (Step 5 Preview Free) — live chat probe returns "Upstream request failed: Model is unavailable", deprecated in the models.dev opencode registry, absent from the docs free table; reverses the 2026-10-10 add (Step 5 stays free on NVIDIA NIM)
+  // Removed (2026-10-17 audit): ling-3.1-flash-free (Ling 3.1 Flash Free) — live chat probe returns "Model is unavailable", deprecated in the models.dev registry, absent from the docs free table
+  // Removed (2026-10-17 audit): ling-3.0-flash-fin-free (Ling 3.0 Flash Fin Free) — live chat probe returns "Model is unavailable", deprecated in the models.dev registry, absent from the docs free table
   ['nemotron-3-ultra-free', 'Nemotron 3 Ultra Free', 'S+', '71.9%', '1M'],
   // Removed (2026-09-05): hy3-free (Tencent Hy3 Free) — absent from live /v1/models (66 models checked)
   ['nemotron-3.5-lightning-free', 'Nemotron 3.5 Lightning Free', 'S+', '-', '262k'], // Added (2026-08-13)
   // Removed (2026-09-05): laguna-s-2.1-free (Laguna S 2.1 Free) - deprecated: marked status=deprecated in the models.dev registry (2026-09-05) and absent from both the Zen /v1/models endpoint and the docs free-models list; the limited-time promo ended
-  ['ling-3.0-flash-fin-free', 'Ling 3.0 Flash Fin Free', 'B+', '-', '262k'], // Added (2026-09-05) — new id in live /v1/models (was ling-3.0-flash-free)
-  // Removed (2026-09-29 audit): muse-spark-1.2-contributor-free (Muse Spark 1.2 Contributor Free) — dropped from the docs free-models table (only the 1.3 contributor variant remains); the 2026-09-28 "re-verify next audit" verdict
-  ['muse-spark-1.3-contributor-free', 'Muse Spark 1.3 Contributor Free', 'S+', '-', '1M'],
+  // Removed (2026-10-17 audit): muse-spark-1.3-contributor-free (Muse Spark 1.3 Contributor Free) — gateway returns an explicit ModelDeprecated error; the contributor promo ended and only the paid muse-spark-1.3 ($1.25/$4.25 per 1M) remains
   ['jev-1.13-free', 'Jev 1.13 Free', 'B+', '-', '200k'], // Added (2026-09-21) — new free model on the live Zen /v1/models list (74 models checked)
   ['space-bunny-free', 'Space Bunny Free', 'S+', '-', '1M'], // Fixed (2026-09-29 audit): ctx '200k' → '1M' (models.dev opencode registry: 1048576)
   ['longcat-2.5-preview-free', 'LongCat 2.5 Preview Free', 'A+', '-', '1M'], // Fixed (2026-09-29 audit): ctx '200k' → '1M' (models.dev opencode registry: 1000000)
+  // Added (2026-10-17 audit) — new free promo wave on the live Zen list (verified via FreeTierError chat probes + models.dev; note /v1/models omits some still-live free ids):
+  ['beam-free', 'Beam Free', 'S+', '80.9%', '1M'], // Added (2026-10-17 audit) — Reflection Beam 1.0 (501B MoE), SWE-bench Verified 80.9% per reflection.ai; also free on Kilo/Vercel/Ollama
+  ['exo-free', 'Exo Free', 'S+', '-', '1M'], // Added (2026-10-17 audit) — new free flagship slot, 1M ctx per models.dev
+  ['nemotron-3.5-super-vl-free', 'Nemotron 3.5 Super VL Free', 'S+', '71.1%', '1M'], // Added (2026-10-17 audit) — 1M ctx on Zen; SWE score follows the same model on NVIDIA NIM, S+ on the documented scale
+  ['deepseek-v4.1-flash-free', 'DeepSeek V4.1 Flash Free', 'S+', '-', '200k'], // Added (2026-10-17 audit) — capped at 200k on Zen
+  ['glm-5.3-flash-free', 'GLM 5.3 Flash Free', 'A+', '-', '200k'], // Added (2026-10-17 audit) — capped at 200k on Zen; tier follows the GLM-5.3-Flash reads on other gateways
   // 2026-09-29 audit: the free-tier gate hardened since issue #181 — 403 FreeTierError even WITH the x-opencode-session header, only space-bunny-free answers plain requests; if TUI probes fail, use a real Zen key
 ]
@@ -479,24 +494,27 @@ export const kilo = [
   ['kilo-auto/free',                         'Kilo Auto Free',      'A+', '-',     '256k'],
   // Removed (2026-09-05): kilo-auto/small (Kilo Auto Small) - no longer free: gateway now lists it with isFree=false and paid pricing ($0.05/M prompt, $0.40/M completion); it routes to paid small models
   ['thinkingmachines/inkling-small:free', 'Thinking Machines Inkling Small (free)', 'S+', '80.2%', '1M'], // tier fixed: 80.2% >= 70% is S+ on the documented scale
-  ['stepfun/step-5-preview-free', 'StepFun Step 5 Preview (free)', 'S', '-', '1M'], // Added (2026-10-10 audit) — new free flagship on the live gateway (replaces the retired step-3.7:free slot); agentic coding, 1M ctx
+  ['reflection/beam-free', 'Reflection Beam 1.0 (free)', 'S+', '80.9%', '1M'], // Added (2026-10-17 audit) — new free flagship on the live gateway (501B MoE, released ~2026-10-05), SWE-bench Verified 80.9% per reflection.ai
+  // Removed (2026-10-17 audit): stepfun/step-5-preview-free (StepFun Step 5 Preview (free)) — free endpoint delisted from the gateway, stepfun/step-5-preview now lists isFree:false with per-token pricing; reverses the 2026-10-10 add (Step 5 stays free on NVIDIA NIM)
   // Removed (2026-10-10 audit): stepfun/step-3.7-flash:free — :free variant gone from the gateway (390 models checked), only paid step-3.7-flash remains; replacement: stepfun/step-5-preview-free
   ['poolside/laguna-s-2.1:free', 'Poolside Laguna S 2.1 (free)', 'A+', '-', '262k'],
   ['nvidia/nemotron-3-ultra-550b-a55b:free', 'NVIDIA Nemotron 3 Ultra (free)', 'A+', '-', '1M'],
+  ['nvidia/nemotron-3.5-super-vl:free', 'NVIDIA Nemotron 3.5 Super VL (free)', 'S+', '71.1%', '1M'], // Added (2026-10-17 audit) — new $0 entry on the live gateway, 1M ctx; SWE score follows the same model on NVIDIA NIM, S+ on the documented scale
+  ['xingchen-agi/xing4.0-29b-a4b:free', 'XingChen Xing4.0 29B (free)', 'S+', '75.0%', '262k'], // Added (2026-10-17 audit) — new $0 entry on the live gateway; SWE-bench Verified 75.0% per the TeleAI model card (same model is the $0 pick on SiliconFlow/OpenRouter)
   // Removed (2026-09-15): minimax/minimax-m2.7:free (MiniMax M2.7 (free)) — no longer free on Kilo gateway
   ['cohere/north-mini-code:free', 'Cohere North Mini Code (free)', 'A-', '-', '256k'],
   ['nvidia/nemotron-3-super-120b-a12b:free', 'NVIDIA Nemotron 3 Super (free)', 'A-', '-', '262k'],
   ['poolside/laguna-xs-2.1:free', 'Poolside Laguna XS 2.1 (free)', 'B+', '-', '262k'],
   ['nvidia/nemotron-3.5-lightning:free', 'NVIDIA Nemotron 3.5 Lightning (free)', 'B+', '-', '1M'],
   ['dots-studio/dots-3-note-preview:free', 'Dots Studio Dots3-Note Preview (free)', 'B+', '-', '512k'],
-  ['openrouter/free', 'OpenRouter Free Models Router', 'B', '-', '-'], // Fixed (2026-10-10 audit): ctx '200k' → '-' (live gateway now returns no context_length for the router)
+  ['openrouter/free', 'OpenRouter Free Models Router', 'B', '-', '200k'], // Fixed (2026-10-17 audit): ctx '-' → '200k' (live gateway metadata exposes context_length again; reverses the 2026-10-10 read)
   // Removed (2026-09-15): minimax/minimax-m3:free (MiniMax M3 (free)) — free variant removed from gateway
   // Removed (2026-09-15): thinkingmachines/inkling:free (Inkling (free)) — free variant no longer exposed
   ['liquid/lfm-2.5-2.6b:free', 'LFM2.5-2.6B (free)', 'C', '-', '64k'], // Added (2026-09-15) — verified via live audit
   // Removed (2026-09-28 audit): z-ai/glm-5.2:free (GLM 5.2 (free)) — no longer free on Kilo: now billed $1.40/$4.40 per M
   // Removed (2026-09-29 audit): inclusionai/ling-3.0-flash-fin:free (Ling 3.0 Flash Fin (free)) — :free variant gone from the catalog (395 models checked), id survives only as paid ($0.075/M prompt); replacement: ling-3.0-flash-sante:free
-  // Removed (2026-10-10 audit): inclusionai/ling-3.0-flash-sante:free — :free variant gone from the gateway (only paid remains), and it is a health/medicine-specialized model anyway; replacement: inclusionai/ling-3.1-flash
-  ['inclusionai/ling-3.1-flash', 'Ling 3.1 Flash (free)', 'A-', '-', '262k'], // Added (2026-10-10 audit) — new $0 entry on the live gateway (560B-A25B MoE); replaces the ling-3.0 :free slots
+  // Removed (2026-10-10 audit): inclusionai/ling-3.0-flash-sante:free — :free variant gone from the gateway (only paid remains), and it is a health/medicine-specialized model anyway
+  // Removed (2026-10-17 audit): inclusionai/ling-3.1-flash (Ling 3.1 Flash (free)) — listing now shows isFree:false with paid per-token pricing, no free variant; reverses the 2026-10-10 add
   // Removed (2026-09-28 audit): inclusionai/ling-3.0-flash-vl:free — :free variant removed from the Kilo catalog (base now billed)
   // Removed (2026-09-28 audit): nex-agi/nex-n2.5-mini:free + nex-agi/nex-n2.5-pro:free — nex-agi provider dropped from the Kilo catalog entirely
   ['nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free', 'NVIDIA Nemotron 3 Nano Omni (free)', 'A+', '-', '256k'], // Fixed (2026-09-28 audit): ctx '262k' → '256k', // Added (2026-09-21) — new in the live free gateway list
@@ -521,9 +539,10 @@ export const llm7 = [
   // ── A+ tier — SWE-bench Verified 50–60% ──
   // Removed (2026-09-05): gemini-3.1-flash-lite (Gemini 3.1 Flash Lite) — now tier=pro usage_based_only (paid) per live /v1/models
   ['kimi-k2.7-code', 'Kimi K2.7 Code', 'A', '-', '256k'], // Added (2026-10-10 audit) — code-tuned Kimi variant, turbo tier usage_based_only=false, 100% availability; score follows the K2.7 Code family (60.4% on ollama-cloud) pending a live reading
-  ['kimi-k2.6', 'Kimi K2.6', 'A', '-', '262k'], // Added (2026-10-10 audit) — turbo tier, usage_based_only=false, tools, 100% availability; tier follows the family A-class reads elsewhere
+  ['kimi-k2.6', 'Kimi K2.6', 'A', '-', '256k'], // Fixed (2026-10-17 audit): ctx '262k' → '256k' (live metadata); added 2026-10-10 — turbo tier, usage_based_only=false, tools
   ['mistral-Nemo-Instruct-2407', 'Mistral Nemo 12B Instruct', 'A-', '-', '128k'], // Added (2026-08-13)
   ['gpt-oss:20b', 'GPT-OSS 20B', 'B+', '-', '128k'], // Added (2026-10-10 audit) — open-weights 20B reasoning, turbo tier usage_based_only=false
+  ['minimax-m2.7', 'MiniMax M2.7', 'S+', '78.0%', '180k'], // Re-added (2026-10-17 audit) — back on the live /v1/models as turbo tier (free), 180k ctx; tier + score follow the same model on routeway/ollama-cloud. Reverses the 2026-10-10 removal
   ['nemotron-3-nano:30b', 'Nemotron 3 Nano 30B', 'B+', '-', '1M'], // Added (2026-10-10 audit) — turbo tier, 1M ctx, 100% availability; score follows the same model on ollama-cloud (38.8%)
   // ── A tier — SWE-bench Verified 40–50% ──
   ['codestral-latest', 'Codestral Latest', 'A', '40.0%', '32k'],
@@ -567,19 +586,22 @@ export const routeway = [
 // 📖 Novita is mostly paid/trial-credit, so this catalog only includes live chat models reporting 0 input/output price.
 // 📖 Test/dev/placeholder zero-price IDs were intentionally excluded.
 export const novita = [
-  // 📖 2026-09-29 audit: the zero-price tier is BACK (live api.novita.ai/v3/openai/models, 119 models) — it had emptied out by 2026-08-13, Novita restored free entries late September
+  // 📖 2026-10-17 audit: the zero-price tier collapsed again — 7 of 8 catalog models now bill per
+  // 📖 token on the live /v1/models (the Ling TIME LIMITED FREE promo and the GLM/Qwen/MiniMax free
+  // 📖 tiers all ended). Only dev/glm46 ($0, 256k) + the new apodex partner launch remain free.
   // ── S+ tier — SWE-bench Verified ≥70% ──
   // Removed (2026-10-10 audit): qwen/qwen3.6-plus + qwen/qwen3.5-plus — both gone from the live /v1/models (94 models): Novita killed the Qwen3.5/3.6 Plus tier in favor of the Qwen3.8 generation; replacements: qwen/qwen3.8-flash (cheap, 1M ctx) / qwen/qwen3.8-max (flagship)
   // ── A tier — SWE-bench Verified 40–50% ──
-  ['zai-org/glm-5.3-flash', 'GLM 5.3 Flash', 'A', '-', '1M'], // Added (2026-10-10 audit) — live on /v1/models + official catalog page; tier follows the GLM-5.3-Flash A-class reads on onomeo/vercel
-  ['qwen/qwen3.8-flash', 'Qwen3.8 Flash', 'A-', '-', '1M'], // Added (2026-10-10 audit) — Qwen3.8 gen replacement for the removed Plus tier; tier follows qwen3.8-flash on DashScope (A+, re-tiered down for Novita's trial-credit reality)
+  // Removed (2026-10-17 audit): zai-org/glm-5.3-flash (GLM 5.3 Flash) — free promo ended, API now bills $1.5/M in + $5/M out; reverses the 2026-10-10 add
+  // Removed (2026-10-17 audit): qwen/qwen3.8-flash (Qwen3.8 Flash) — no longer free, now bills $1.5/M in + $4.7/M out; reverses the 2026-10-10 add
   // Removed (2026-09-29 audit): inclusionai/ling-3.0-flash-fin (Ling 3.0 Flash Fin) — still listed but now billed $0.075/M in, $0.22/M out; replacement: ling-3.1-flash
-  ['inclusionai/ling-3.0-flash-sante', 'Ling 3.0 Flash Sante', 'B+', '-', '262k'], // Fixed (2026-09-29 audit): ctx '256k' → '262k' (live context_size field); only Ling 3.0 variant still zero-price
-  ['inclusionai/ling-3.1-flash', 'Ling 3.1 Flash', 'A', '-', '262k'], // Added (2026-09-29 audit) — new (~Sep 26) zero-price model; API description claims 1M ctx but context_size enforces 262k
-  ['dev/glm46', 'GLM 4.6 (dev)', 'A', '-', '256k'], // Re-added (2026-09-29 audit) — zero-price again on the live API; reverses the 2026-09-22 paid-only removal
-  ['inclusionai/ling-3.0-flash', 'Ling 3.0 Flash', 'B+', '-', '262k'], // Added (2026-10-10 audit) — new on the live /v1/models, time-limited free promo
-  ['zai-org/glm-4.7-flash', 'GLM 4.7 Flash', 'B+', '-', '200k'], // Added (2026-10-10 audit) — live on /v1/models; tier follows @cf/zai-org/glm-4.7-flash on cloudflare
-  ['minimax/minimax-m2.7', 'MiniMax M2.7', 'A-', '-', '200k'], // Added (2026-10-10 audit) — live on /v1/models; tier follows the M2.7 A-class read on onomeo
+  // Removed (2026-10-17 audit): inclusionai/ling-3.0-flash-sante (Ling 3.0 Flash Sante) — TIME LIMITED FREE promo ended, now $0.75/M in + $2.2/M out
+  // Removed (2026-10-17 audit): inclusionai/ling-3.1-flash (Ling 3.1 Flash) — no longer free, now bills $3/M in + $9/M out; reverses the 2026-09-29 add
+  ['dev/glm46', 'GLM 4.6 (dev)', 'A', '-', '256k'], // Re-added (2026-09-29 audit) — zero-price again on the live API; reverses the 2026-09-22 paid-only removal; 2026-10-17 audit: still the only $0 legacy model
+  ['apodex/apodex-1.1-mini', 'Apodex 1.1 Mini', 'B+', '-', '262k'], // Added (2026-10-17 audit) — free partner launch since 2026-10-01 (35B MoE, no stated end date), $0 on the live API; replaces its OpenRouter :free slot delisted the same day
+  // Removed (2026-10-17 audit): inclusionai/ling-3.0-flash (Ling 3.0 Flash) — no longer free, now bills $0.6/M in + $1.8/M out; reverses the 2026-10-10 add
+  // Removed (2026-10-17 audit): zai-org/glm-4.7-flash (GLM 4.7 Flash) — no longer free, now bills $0.7/M in + $4/M out; reverses the 2026-10-10 add
+  // Removed (2026-10-17 audit): minimax/minimax-m2.7 (MiniMax M2.7) — no longer free, now bills $3/M in + $12/M out; reverses the 2026-10-10 add
   // Notes (2026-09-29): `bunny` is also zero-price again but Novita publishes no description, coding relevance unverifiable; left out
 ]
@@ -615,7 +637,7 @@ export const pollinations = [
   // Removed (2026-09-28 audit): anthropic/claude-opus-5 — now paid_only in the Pollinations catalog
   // ── S tier — SWE-bench Verified 60–70% ──
   // Removed (2026-09-28 audit): anthropic/claude-sonnet-5 — now paid_only in the Pollinations catalog
-  ['qwen-coder', 'Qwen3 Coder 30B', 'A+', '51.6%', '262k'], // Fixed (2026-09-21): alias now resolves to qwen/qwen3-coder-30b-a3b-instruct; re-scored from the 480B figure to the 30B SWE-bench Verified
+  // Removed (2026-10-17 audit): qwen-coder (Qwen3 Coder 30B) — flipped to paid: qwen/qwen3-coder-30b-a3b-instruct now carries paid_only=true and anonymous calls 401; still free on Scaleway
   ['deepseek', 'DeepSeek V4 Flash', 'S+', '79.0%', '1M'], // Fixed (2026-09-21): alias now resolves to deepseek/deepseek-v4-flash (V4 Flash 0731), was V3; re-scored per the V4 Flash family entry
   // Removed (2026-09-28 audit): kimi-code (Kimi K2.7 Code) — now paid_only in the Pollinations catalog
   ['openai', 'OpenAI GPT-5.4 Nano', 'B+', '-', '400k'], // Fixed (2026-09-21): alias now resolves to openai/gpt-5.4-nano (was a generic GPT alias); re-tiered to the nano class
@@ -625,23 +647,30 @@ export const pollinations = [
   ['deepseek/deepseek-v4.1-flash', 'DeepSeek V4.1 Flash', 'A+', '-', '1M'], // Added (2026-09-22) — healthy on live /v1/models, successor to V4 Flash
   // Removed (2026-09-28 audit): meituan/longcat-2.0 (LongCat 2.0) — now paid_only in the Pollinations catalog
   // Removed (2026-09-26): gemma-4-31b (Gemma 4 31B) — requires paid Pollen (embedded credits error on free-tier keys, issue #190); the free Gemma route is NVIDIA NIM
-  ['gpt-oss', 'GPT OSS 20B', 'A+', '50.3%', '131k'],
+  ['gpt-oss', 'GPT OSS 20B', 'A+', '50.3%', '128k'], // Fixed (2026-10-17 audit): ctx '131k' → '128k' (live gen.pollinations.ai metadata)
   // Removed (2026-09-28 audit): qwen3.7-flash — now paid_only in the Pollinations catalog; still free on DashScope
   // Added (2026-09-28 audit) — 7 new free models on the live gen.pollinations.ai catalog:
   ['openai/gpt-6-sol', 'OpenAI GPT-6 Sol', 'S+', '-', '1M'],
   ['openai/gpt-5.3-codex', 'OpenAI GPT-5.3 Codex', 'S+', '-', '400k'],
   ['openai/gpt-6-luna', 'OpenAI GPT-6 Luna', 'S', '-', '1M'],
-  ['z-ai/glm-5.2', 'Z.ai GLM-5.2', 'S', '-', '1M'],
+  // Removed (2026-10-17 audit): z-ai/glm-5.2 — flipped to paid (paid_only=true, anonymous 401); reverses the 2026-09-28 add
   ['openai/gpt-5.4-mini', 'OpenAI GPT-5.4 Mini', 'A+', '-', '400k'],
-  ['x-ai/grok-4.20', 'xAI Grok 4.20', 'A+', '-', '262k'],
-  ['amazon/nova-2-lite-v1', 'Amazon Nova 2 Lite', 'B+', '-', '1M'], // Fixed (2026-09-29 audit): id 'amazon/nova-2-lite' → 'amazon/nova-2-lite-v1' (delisted under the old id, live under the v1 id, same 1M ctx, healthy)
+  ['x-ai/grok-4.20', 'xAI Grok 4.20', 'A+', '-', '256k'], // Fixed (2026-10-17 audit): ctx '262k' → '256k' (live catalog metadata)
+  // Removed (2026-10-17 audit): amazon/nova-2-lite-v1 — flipped to paid (paid_only=true, anonymous 401)
+  // Removed (2026-10-17 audit): x-ai/grok-4.7 — flipped to paid (paid_only=true, anonymous 401); reverses the 2026-10-10 add, grok-4.6 is the free successor
+  // Removed (2026-10-17 audit): stepfun/step-5-preview — flipped to paid (paid_only=true, anonymous 401); reverses the 2026-10-10 add (Step 5 stays free on NVIDIA NIM)
   // Added (2026-10-10 audit) — 3 new free models on live gen.pollinations.ai/v1/models (the 4 other candidates found the same day were skipped: kimi-k2.7-code, qwen3-coder-next, laguna-s-2.1, longcat-2.0 all carry a 2026-09 "paid Pollen" verdict and the audit did not sniff completions to clear it, see issue #190 note above):
   ['openai/gpt-6.1-sol', 'OpenAI GPT-6.1 Sol', 'S+', '-', '1M'], // Added (2026-10-10 audit) — new in the live catalog, GPT-6 Sol successor, 1M ctx
-  ['x-ai/grok-4.7', 'xAI Grok 4.7', 'S', '-', '500k'], // Added (2026-10-10 audit) — new in the live catalog, Grok 4.20 successor, 500k ctx
-  ['stepfun/step-5-preview', 'StepFun Step 5 Preview', 'A+', '-', '1M'], // Added (2026-10-10 audit) — new in the live catalog (same Step 5 rollout as Kilo/llm7/opencode-zen), 1M ctx
+  // Added (2026-10-17 audit) — 6 new free models on live gen.pollinations.ai/v1/models (5 catalog slots flipped to paid Pollen the same day):
+  ['qwen/qwen3.8-2.4t-a95b', 'Qwen3.8 2.4T A95B', 'A+', '-', '256k'], // Added (2026-10-17 audit) — the DashScope 2.4T flagship, capped at 256k here; tier follows the DashScope entry
+  ['x-ai/grok-4.6', 'xAI Grok 4.6', 'A+', '-', '200k'], // Added (2026-10-17 audit) — free successor of the paid-flipped grok-4.7
+  ['openai/gpt-5.4', 'OpenAI GPT-5.4', 'S', '-', '1M'], // Added (2026-10-17 audit) — full GPT-5.4 at 1M ctx (the 'openai' alias keeps routing to gpt-5.4-nano)
+  ['x-ai/grok-4.3', 'xAI Grok 4.3', 'B+', '-', '200k'], // Added (2026-10-17 audit) — legacy Grok slot back on the free tier
+  ['nvidia/nemotron-3-ultra', 'NVIDIA Nemotron 3 Ultra', 'S+', '71.9%', '256k'], // Re-added (2026-10-17 audit) — back to paid_only=false in the live catalog; reverses the 2026-09-26 paid-Pollen removal, score follows the same model on NVIDIA
+  ['mistralai/mistral-large-3', 'Mistral Large 3', 'A+', '-', '256k'], // Added (2026-10-17 audit) — new free listing; tier follows mistral-large-3:675b on ollama-cloud
   // Audit (2026-09-28): text.pollinations.ai/models endpoint is down (502); catalog now read from gen.pollinations.ai
   // ── B+ tier ──
-  ['nemotron-3.5-lightning', 'Nemotron 3.5 Lightning', 'B+', '-', '262k'],
+  ['nemotron-3.5-lightning', 'Nemotron 3.5 Lightning', 'B+', '-', '256k'], // Fixed (2026-10-17 audit): ctx '262k' → '256k' (live catalog metadata)
 ]
 // 📖 SiliconFlow source - https://api.siliconflow.cn/v1/chat/completions
@@ -676,19 +705,19 @@ export const requesty = [
   ['nvidia/nemotron-3-super-120b-a12b', 'Nemotron 3 Super', 'S', '60.5%', '1M'],
   ['poolside/laguna-m.1', 'Laguna M.1', 'S', '-', '32k'],
   // ── A+ tier — SWE-bench Verified 50–60% ──
-  ['google/gemma-4-31b-it', 'Gemma 4 31B', 'A+', '52.0%', '262k'],
-  ['nvidia/nemotron-3-nano-omni-30b-a3b-reasoning', 'Nemotron 3 Omni', 'A+', '52.0%', '131k'],
+  ['google/gemma-4-31b-it', 'Gemma 4 31B', 'A+', '52.0%', '256k'], // Fixed (2026-10-17 audit): ctx '262k' → '256k' (live context_window on router.requesty.ai)
+  ['nvidia/nemotron-3-nano-omni-30b-a3b-reasoning', 'Nemotron 3 Omni', 'A+', '52.0%', '128k'], // Fixed (2026-10-17 audit): ctx '131k' → '128k' (live context_window)
   // ── A tier — SWE-bench Verified 40–50% ──
-  ['nvidia/nemotron-3-nano-30b-a3b', 'Nemotron Nano 30B', 'A-', '38.8%', '262k'],
+  ['nvidia/nemotron-3-nano-30b-a3b', 'Nemotron Nano 30B', 'A-', '38.8%', '256k'], // Fixed (2026-10-17 audit): ctx '262k' → '256k' (live context_window)
   // ── B+ tier — SWE-bench Verified 30–35% ──
   ['nvidia/nemotron-3.5-lightning-30b-a3b', 'Nemotron 3.5 Lightning', 'B+', '-', '1M'],
   // ── B tier — SWE-bench Verified 20–30% ──
-  ['mistral/leanstral-1-5', 'Leanstral 1.5', 'B', '-', '262k'],
-  ['novita/inclusionai/ling-3.0-tiny', 'Ling 3.0 Tiny', 'B', '-', '262k'],
-  ['novita/inclusionai/ling-3.1-flash', 'Ling 3.1 Flash', 'B+', '-', '262k'], // Added (2026-10-10 audit) — new on the router since ~2026-09-28, $0/$0 per live /v1/models; mirrors novita's ling-3.1-flash
+  ['mistral/leanstral-1-5', 'Leanstral 1.5', 'B', '-', '256k'], // Fixed (2026-10-17 audit): ctx '262k' → '256k' (live context_window)
+  ['novita/inclusionai/ling-3.0-tiny', 'Ling 3.0 Tiny', 'B', '-', '256k'], // Fixed (2026-10-17 audit): ctx '262k' → '256k' (live context_window)
+  ['novita/ling-3.1-flash', 'Ling 3.1 Flash', 'B+', '-', '256k'], // Fixed (2026-10-17 audit): id 'novita/inclusionai/ling-3.1-flash' → 'novita/ling-3.1-flash' (silent router rename, the inclusionai path was dropped) + ctx '262k' → '256k' (live context_window); added 2026-10-10
   // ── C tier — other zero-price models (kept for breadth) ──
   // Removed (2026-10-10 audit): nvidia/nemotron-3.5-content-safety — moderation guardrail classifier (Gemma-3-4B fine-tune, tools:false), not a coding model (catalog hygiene; matches OpenRouter/Kilo cleanup)
-  ['nvidia/muse-glimmer-30b', 'Muse Glimmer 30B', 'C', '-', '131k'],
+  ['nvidia/muse-glimmer-30b', 'Muse Glimmer 30B', 'C', '-', '128k'], // Fixed (2026-10-17 audit): ctx '131k' → '128k' (live context_window)
 ]
 // 📖 OrcaRouter source - https://api.orcarouter.ai/v1
@@ -728,11 +757,14 @@ export const vercelGateway = [
   // Removed (2026-09-15): minimax/minimax-m3-free (MiniMax M3 (Free)) — free variant removed, now paid-only
   // Removed (2026-09-15): minimax/minimax-m2.7-free (MiniMax M2.7 (Free)) — free variant removed, now paid-only
   ['poolside/laguna-s-2.1-free', 'Laguna S 2.1 (Free)', 'S+', '-', '256k'], // tier follows family precedent: laguna-xs-2.1 ships S+ 70.9% via NVIDIA. 2026-09-29 audit: still served $0/$0 live, the models.dev deprecation flag does not apply to the Vercel variant
+  ['reflection/beam-free', 'Beam 1.0 (Free)', 'S+', '80.9%', '1M'], // Added (2026-10-17 audit) — new $0/$0 flagship (Reflection Beam 1.0, 501B MoE, released 2026-10-05), SWE-bench Verified 80.9% per reflection.ai
   // ── B+ tier — vertical-tuned lightweight (coding secondary) ──
   // Removed (2026-09-28 audit): inclusionai/ling-3.0-flash-fin-free + inclusionai/ling-3.0-flash-vl-free — free variants removed from the gateway, only paid variants remain
   // Removed (2026-10-10 audit): inclusionai/ling-3.0-flash-sante-free — the free slot was upgraded: gateway now serves ling-3.1-flash-free at $0/$0 (paid ling-3.0-flash-sante still listed)
-  ['inclusionai/ling-3.1-flash-free', 'Ling 3.1 Flash (Free)', 'B+', '-', '262k'], // Added (2026-10-10 audit) — new $0/$0 free slot in the live gateway (415 models checked)
+  // Removed (2026-10-17 audit): inclusionai/ling-3.1-flash-free — delisted from the gateway entirely, only the paid inclusionai/ling-3.1-flash remains ($0.30/$0.90 per M); reverses the 2026-10-10 add
+  ['nvidia/nemotron-3.5-lightning-free', 'Nemotron 3.5 Lightning (Free)', 'A+', '54.3%', '1M'], // Added (2026-10-17 audit) — new $0/$0 slot; SWE-bench Verified 54.3% per the NVIDIA model card, A+ on the documented scale (matches the NIM entry)
   ['stealth/glyph-cluster', 'Glyph Cluster (stealth)', 'B', '-', '256k'], // Added (2026-10-10 audit) — new anonymous early-access model (~2026-10-07), $0/$0, 256k ctx; replaces the rotated-out pixel-canary slot
+  ['stealth/cat-jam', 'Cat Jam (stealth)', 'B+', '-', '1M'], // Added (2026-10-17 audit) — new stealth free model per the official Vercel changelog, $0/$0, 1M ctx
   // Removed (2026-10-10 audit): stealth/pixel-canary — stealth slot rotated, id absent from the live gateway
   // 2026-09-29 audit: list stays strictly $0/$0. Credit-value alternates for the $5/30d credits (paid, NOT added): zai/glm-5.3-flash ($0.15/$0.50 per M, 1M ctx), deepseek/deepseek-v4-flash ($0.13/$0.26 per M, 1M ctx), qwen3.7-flash ($0.03/$0.13 per M, 991k ctx)
 ]
@@ -744,8 +776,8 @@ export const ollamaCloud = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   ['nemotron-3-ultra', 'Nemotron 3 Ultra', 'S+', '71.9%', '256k'],
   // Removed (2026-09-28 audit): glm-5.1 (GLM 5.1) — no longer in the Ollama Cloud catalog (glm-5.2/5.3/5.3-flash remain)
-  ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '976k'], // Fixed (2026-07-27): ctx '128k' → '1M'
-  ['minimax-m2.7', 'MiniMax M2.7', 'S+', '78.0%', '200k'],
+  ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '1M'], // Fixed (2026-10-17 audit): ctx '976k' → '1M' (live /v1/models reports the full 1048576); ctx first fixed 2026-07-27
+  ['minimax-m2.7', 'MiniMax M2.7', 'S+', '78.0%', '192k'], // Fixed (2026-10-17 audit): ctx '200k' → '192k' (exact live value)
   ['minimax-m3', 'MiniMax M3', 'S+', '78.4%', '512k'], // Fixed (2026-07-27): ctx '512k' → '1M'
   // Removed (2026-08-23): minimax-m2.5 (MiniMax M2.5) — no longer in ollama.com/v1/models (19 models live)
   ['kimi-k2.6', 'Kimi K2.6', 'S+', '80.2%', '256k'], // Fixed (2026-07-27): ctx '262k' → '256k'
@@ -767,14 +799,15 @@ export const ollamaCloud = [
   ['mistral-large-3:675b', 'Mistral Large 3 675B Cloud', 'A+', '-', '256k'], // Fixed (2026-08-23): ID 'mistral-large-3:675b-cloud' → 'mistral-large-3:675b' (tag renamed upstream)
   ['mistral-large-4', 'Mistral Large 4', 'A+', '-', '1M'], // Added (2026-10-10 audit) — new cloud model (~Oct 4, 1.05T MoE, 1M ctx, thinking + tools); tier follows the Mistral Large family precedent
   // Removed (2026-09-28 audit): qwen3.5:397b (Qwen 3.5 Cloud) — qwen3.5 is not flagged cloud on ollama.com and is absent from the cloud catalog/pricing page
-  ['nemotron-3-nano:30b', 'Nemotron 3 Nano 30B', 'A-', '38.8%', '1M'],
+  ['nemotron-3-nano:30b', 'Nemotron 3 Nano 30B', 'A-', '38.8%', '256k'], // Fixed (2026-10-17 audit): ctx '1M' → '256k' (live /v1/models downgraded the context)
+  ['beam', 'Beam', 'S+', '81.6%', '1M'], // Added (2026-10-17 audit) — Reflection Beam free on Ollama Cloud for a limited time per the 2026-10-16 blog; SWE-bench Verified 81.6% per Ollama's post
 ]
 // 📖 onomeo source - https://onomeo.com/docs
 // 📖 OpenAI-compatible gateway: https://onomeo.com/v1/chat/completions (streaming + tool calls)
 // 📖 Public beta: not every feature is guaranteed to work; feedback is welcome at https://onomeo.com/feedback.
 // 📖 The models listed here are free models: they spend no credits and are limited by calls instead.
-// 📖 Limits: 12 req/min per key, 60 calls per 5 hours per account (a long request counts as several),
+// 📖 Limits: 12 req/min per key, 30 calls per 5 hours per account (a long request counts as several, reduced from 60 in Oct 2026),
 // 📖 120 per 5 hours per IP, and a site-wide pool of 450 per 5 hours shared by all accounts.
 // 📖 Premium models are not listed; they spend credits from a daily check-in (no card: 20,000 on day 1,
 // 📖 rising to 50,000/day from day 7 of a streak), and each account that has not paid can spend up to
@@ -786,7 +819,7 @@ export const onomeo = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   ['deepseek-v4-flash', 'DeepSeek V4 Flash', 'S+', '79.0%', '1M'],
   ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '1M'],
-  ['gemini-3.8-flash', 'Gemini 3.8 Flash', 'S+', '-', '1M'], // Added (2026-09-29 audit) — in the free pool since 2026-09-23; tier follows the same model on googleai
+  // Removed (2026-10-17 audit): gemini-3.8-flash (Gemini 3.8 Flash) — delisted, absent from the current models page (neither free nor paid listing); gemini-3.7-flash remains the free Gemini slot
   ['gemini-3.7-flash', 'Gemini 3.7 Flash', 'S+', '-', '1M'], // Added (2026-09-29 audit) — in the free pool since 2026-09-23; tier follows the same model on googleai; 2026-10-10 audit: still listed but upstream-paused
   ['gemini-3.6-flash', 'Gemini 3.6 Flash', 'S+', '-', '1M'], // Added (2026-10-10 audit) — new free model on the live catalog (model page); tier follows the same model on googleai
   ['gemini-3.5-flash', 'Gemini 3.5 Flash', 'S+', '78.0%', '1M'], // Added (2026-09-29 audit) — in the free pool since 2026-09-23; score from the same model on googleai
@@ -796,18 +829,19 @@ export const onomeo = [
   // Removed (2026-10-10 audit): nvidia/nemotron-3-super-120b-a12b (Nemotron 3 Super) — in the provider's own removed list
   // ── A+ tier — SWE-bench Verified 50–60% ──
   ['gpt-oss-20b', 'GPT OSS 20B', 'A+', '50.3%', '131k'], // Fixed (2026-10-10 audit): id 'openai/gpt-oss-20b' → 'gpt-oss-20b' (first-party id rename in the live catalog) + ctx '128k' → '131k'
-  ['glm-5.3-flash', 'GLM-5.3 Flash', 'A+', '-', '1M'], // Fixed (2026-10-10 audit): id 'z-ai/glm-5.3-flash-free' → 'glm-5.3-flash' (first-party id rename in the live catalog)
-  // Removed (2026-10-10 audit): stepfun/step-3.7-flash:free (Step 3.7 Flash) — in the provider's removed list; no step-5-preview on onomeo either
-  ['glm-4.7-flash', 'GLM 4.7 Flash', 'A+', '59.2%', '200k'], // Added (2026-09-29 audit) — in the free pool since 2026-09-28; score from the same Zhipu model on cloudflare (@cf/zai-org/glm-4.7-flash); 2026-10-10 audit: still listed but upstream-paused
+  // Removed (2026-10-17 audit): glm-5.3-flash (GLM-5.3 Flash) — officially removed this week per the site's own changelog (the paid glm-5.3 remains); free GLM-5.3-Flash lives on NVIDIA NIM / Pollinations / OrcaRouter
+  // Removed (2026-10-17 audit): stepfun/step-3.7-flash:free (Step 3.7 Flash) — in the provider's removed list; no step-5-preview on onomeo either
+  // Removed (2026-10-17 audit): glm-4.7-flash (GLM 4.7 Flash) — delisted, absent from the current models page; free route is @cf/zai-org/glm-4.7-flash on Cloudflare
   ['gemini-3.5-flash-lite', 'Gemini 3.5 Flash Lite', 'S', '-', '1M'], // Added (2026-09-29 audit) — in the free pool since 2026-09-23; tier follows the same model on googleai
   // ── A tier — SWE-bench Verified 40–50% ──
   ['codestral-latest', 'Codestral Latest', 'A', '40.0%', '256k'],
   ['glm-4.5-flash', 'GLM 4.5 Flash', 'A', '-', '131k'], // Added (2026-09-29 audit) — in the free pool since 2026-09-28
-  ['glm-4.6v-flash', 'GLM 4.6V Flash', 'A', '-', '128k'], // Added (2026-09-29 audit) — in the free pool since 2026-09-28; vision model; 2026-10-10 audit: still listed but upstream-paused
+  // Removed (2026-10-17 audit): glm-4.6v-flash (GLM 4.6V Flash) — delisted, absent from the current models page; no other free GLM vision model offered
   ['ministral-14b-latest', 'Ministral 3 14B', 'A', '-', '262k'], // Added (2026-09-29 audit) — in the free pool since 2026-09-23
   ['muse-glimmer-30b', 'Muse Glimmer 30B', 'A', '-', '131k'], // Fixed (2026-10-10 audit): id 'meta/muse-glimmer-30b' → 'muse-glimmer-30b' (first-party rename) + ctx '128k' → '131k'
   ['minimax-m2.7', 'MiniMax M2.7', 'A+', '-', '205k'], // Fixed (2026-10-10 audit): ctx '200k' → '205k' (live catalog); still listed but upstream-paused
-  ['stealth/space-bunny-alpha', 'Space Bunny Alpha', 'A', '-', '1M'],
+  ['stealth/glyph-cluster', 'Glyph Cluster', 'A+', '-', '-'], // Added (2026-10-17 audit) — new stealth free model replacing the delisted space-bunny-alpha; tier follows the same stealth model on kilo/vercel-gateway, ctx unpublished on onomeo
+  // Removed (2026-10-17 audit): stealth/space-bunny-alpha — delisted, replaced by the new stealth/glyph-cluster slot
   // Removed (2026-10-10 audit): inclusionai/ling-3.0-flash-sante:free + inclusionai/ling-3.0-flash-fin:free — sante in the provider's removed list (successor ling-3.1-flash also removed upstream), fin absent from the live catalog entirely
   // Removed (2026-10-10 audit): dots-studio/dots-3-note-preview:free + sensenova-6.8-flash-lite + gemini-3-flash-preview + mimo-v2.5:free — dots-3/gemini-3-flash-preview/mimo in the provider's removed list (gemini superseded by the 3.5/3.6/3.8 flash family), sensenova zero occurrences in the live catalog payload
   // Removed (2026-10-10 audit): nex-agi/nex-n2.5-mini:free + nex-agi/nex-n2.5-pro:free + Shanghai_AI_Laboratory/Intern-S2-Preview — all three in the provider's removed list (they were the modelsPaused re-verify set of the 2026-09-29 audit: verdict is now removal)
@@ -844,8 +878,8 @@ export const sources = {
   cerebras: {
     name: 'Cerebras',
     url: 'https://api.cerebras.ai/v1/chat/completions',
-    quota: 'Free · generous dev tier',
-    quotaCode: 'free',
+    quota: 'Free trial · $5 credits / 30 days · card required',
+    quotaCode: 'limited',
     models: cerebras,
   },
   googleai: {
@@ -907,13 +941,16 @@ export const sources = {
     quotaCode: 'free',
     models: codestral,
   },
-  zai: {
-    name: 'ZAI',
-    url: 'https://api.z.ai/api/coding/paas/v4/chat/completions',
-    quota: 'Free · Flash models only',
-    quotaCode: 'free',
-    models: zai,
-  },
+  // 'zai': REMOVED 2026-10-17 — GLM Coding Plan free window ended 2026-10-07, plan is
+  // paid-only now (no free model left on the coding endpoint). Provider metadata and
+  // ZAI_API_KEY support stay in src/core for backwards compat with existing configs.
+  // 'zai': {
+  //   name: 'ZAI',
+  //   url: 'https://api.z.ai/api/coding/paas/v4/chat/completions',
+  //   quota: 'Free · Flash models only',
+  //   quotaCode: 'free',
+  //   models: zai,
+  // },
   scaleway: {
     name: 'Scaleway',
     url: 'https://api.scaleway.ai/v1/chat/completions',
@@ -963,7 +1000,7 @@ export const sources = {
   novita: {
     name: 'Novita AI',
     url: 'https://api.novita.ai/openai/v1/chat/completions',
-    quota: 'Free tier is back · 5 zero-price models · 2026-09-29',
+    quota: 'Free tier shrank · 2 zero-price models left',
     quotaCode: 'limited',
     models: novita,
   },
@@ -1012,7 +1049,7 @@ export const sources = {
   onomeo: {
     name: 'onomeo',
     url: 'https://onomeo.com/v1/chat/completions',
-    quota: 'Free models, no credits · 12 RPM · 60 req/5h',
+    quota: 'Free models, no credits · 12 RPM · 30 req/5h',
     quotaCode: 'limited',
     models: onomeo,
   },