free-coding-models 0.5.63 → 0.5.65

package/README.md

@@ -6,7 +6,7 @@
 
 <p align="center">
   <strong>Find the fastest free coding model in seconds</strong><br>
-  Track ~191 models across 20 trusted free or free-limited AI providers in real time<br><br>
+  Track ~222 models across 20 trusted free or free-limited AI providers in real time<br><br>
   <strong>Install Free API endpoints to your favorite AI coding tools:</strong><br>
   OpenCode CLI / Desktop / WebUI, OpenClaw, Crush, Goose, Aider, Kilo CLI, Qwen Code, OpenHands, Amp, Hermes, Continue, Cline, Xcode, Pi, ZCode and more...<br><br>
   <strong>Use Kimi K2, DeepSeek V3, GPT-OSS, Qwen3, MiniMax M3, GLM, Llama 4, Gemma 4, Devstral and more — for free</strong>
@@ -16,7 +16,7 @@
   <img src="https://img.shields.io/npm/v/free-coding-models?color=3d6b00&label=npm&logo=npm" alt="npm version" width="200"><br>
   <img src="https://img.shields.io/node/v/free-coding-models?color=3d6b00&logo=node.js" alt="node version" width="200"><br>
   <img src="https://img.shields.io/npm/l/free-coding-models?color=3d6b00" alt="license" width="200"><br>
-  <img src="https://img.shields.io/badge/models-191-3d6b00?logo=nvidia" alt="models count" width="200"><br>
+  <img src="https://img.shields.io/badge/models-222-3d6b00?logo=nvidia" alt="models count" width="200"><br>
   <img src="https://img.shields.io/badge/providers-20-1a56db" alt="providers count" width="200">
 </p>
 
@@ -59,7 +59,7 @@ free-coding-models
 
 ## 💡 Why this tool?
 
-There are **~191 cataloged free or free-limited coding models** across 20 vetted providers. Which one is fastest right now? Which one is actually stable versus just lucky on the last ping?
+There are **~222 cataloged free or free-limited coding models** across 20 vetted providers. Which one is fastest right now? Which one is actually stable versus just lucky on the last ping?
 
 This CLI pings them all in parallel, shows live latency, and calculates a **live Stability Score (0-100)**. Average latency alone is misleading if a model randomly spikes to 6 seconds; the stability score measures true reliability by combining **p95 latency** (30%), **jitter/variance** (30%), **spike rate** (20%), and **uptime** (20%). 
 
@@ -73,7 +73,7 @@ It then writes the model you pick directly into your coding tool's config — so
 
 Create a free account on one provider below to get started. A few providers (`Kilo`, `LLM7`, OVHcloud sandbox) can also answer without a key, with tighter shared limits.
 
-**~191 coding models** across 20 active providers, ranked by practical free-tier usefulness.
+**~222 coding models** across 20 active providers, ranked by practical free-tier usefulness.
 
 | # | Provider | Models | Tier range | Free tier | Env var |
 |---|----------|--------|-----------|-----------|--------|
@@ -102,6 +102,17 @@ Create a free account on one provider below to get started. A few providers (`Ki
 
 > 🧹 Audit cleanup: `iFlow` was removed because it shut down on April 17, 2026. `Together AI`, `Perplexity API`, `DeepInfra`, `Replicate`, `Fireworks`, `Hyperbolic`, `Hugging Face`, `SiliconFlow`, `Chutes AI` were removed from the active free catalog because they are paid, trial-credit only, too tiny to be useful, unclear as a stable free API, or tool-specific rather than a generally usable free provider. `Rovo` and `Gemini CLI` were also wiped out as tool integrations (CLI-only, not generally usable free providers).
 
+### ⚠️ Health checks consume provider quota
+
+> FCM continuously health-probes every model in the catalog so you see live latency, status, and verdict. When you've configured a provider's API key (e.g. `OPENROUTER_API_KEY`), **those probes are authenticated and count against that provider's daily quota**. Rate-limited providers like **OpenRouter** (50 req/day free, 1000 req/day at $10 spend) are particularly sensitive: a single overloaded model can burn your quota in minutes if it's re-pinged every second.
+
+> As of v0.3.83 (issue #146), FCM now:
+> - **Auto-pauses** a provider when its probe gets a `429` response (respects `Retry-After` header + the "try again N seconds later" message body).
+> - **Backs off exponentially** on per-model failures (30s → 1m → 2m → 5m) instead of re-pinging a broken model every cycle.
+> - **Surfaces a footer chip** like `⏸ openrouter 14h` so you can see which providers are resting.
+
+> **If you don't need authenticated probes for a provider**, just leave its key empty — anonymous probes still work for liveness checks on most providers, and they won't burn your personal quota.
+
 ---
 
 ### Tier scale
@@ -503,7 +514,7 @@ Inspect or wipe it manually any time — it's plain JSON with `0600` perms.
 ```json
 {
   "probeCache": {
-    "total": 191,
+    "total": 222,
     "ok": 178,
     "broken": 13,
     "freshCount": 165,
@@ -764,7 +775,7 @@ See [`packages/fcm-agent-core/README.md`](./packages/fcm-agent-core/README.md) f
 
 ## ✨ Features
 
-- **Parallel pings** — all ~191 API/Zen-callable models tested simultaneously via native `fetch`
+- **Parallel pings** — all ~222 API/Zen-callable models tested simultaneously via native `fetch`
 - **AI benchmark columns** — `Ctrl+A` benchmarks the selected model, `Ctrl+U` benchmarks visible models, and results split cleanly into **AI Latency** plus **TPS**. Settings includes an opt-in **Startup AI Speed Scan** toggle to run the global benchmark automatically after launch.
 - **Tiny verdict indicator** — the first `❔` column mirrors the full Verdict as a compact emoji (`🟩`, `🟢`, `🟡`, `🟠`, etc.) and sorts by the same verdict order.
 - **Adaptive monitoring** — 2s burst for 60s → 10s normal → 30s idle

package/changelog/v0.5.64.md

@@ -0,0 +1,64 @@
+# Changelog v0.5.64 - 2026-07-27
+
+### Added
+
+- 🔍 **Full provider audit** — every one of the 20 active providers was re-verified against live API endpoints and official docs. Researchers fetched current catalogs and reported structured JSON diffs (`removed` / `added` / `fixed` / `confirmed`). The agent then applied all safe changes to `sources.js` and `audit_state.json`.
+
+  - **8 new free-tier models added** across 7 providers:
+    - `opencode-zen` — `poolside/laguna-s-2.1-free` (Laguna S 2.1 Free), `ling-3.0-flash-free` (Ling-3.0-flash Free)
+    - `openrouter` — `inclusionai/ling-3.0-flash:free` (Ling-3.0 Flash), `poolside/laguna-s-2.1:free` (Poolside Laguna S 2.1), `openai/gpt-oss-20b:free` (GPT OSS 20B)
+    - `googleai` — `gemini-3.6-flash` (Gemini 3.6 Flash), `gemini-3.5-flash-lite` (Gemini 3.5 Flash Lite)
+    - `qwen` — `qwen3.7-flash` (Qwen3.7 Flash)
+    - `scaleway` — `gemma-4-31b-it` (Gemma 4 31B IT)
+    - `llm7` — `gemini-3.1-flash-lite` (Gemini 3.1 Flash Lite)
+    - `ollama-cloud` — `kimi-k3` (Kimi K3), `gemini-3-flash-preview` (Gemini 3 Flash Preview), `mistral-large-3:675b-cloud` (Mistral Large 3 675B Cloud), `qwen3.5` (Qwen 3.5 Cloud)
+
+- 🧪 **New `audit_state.json` fingerprint system** — every provider audit now records `lastAudited` + an MD5 fingerprint of its confirmed + added model IDs. Next audit skips a provider automatically when its fingerprint hasn't changed.
+
+- 📜 **Migrated `provider_updates/` archival to git history** — the v1 skill kept per-provider markdown reports + per-version archives. v2 keeps git as the single source of truth: every `sources.js` change ships in a commit with a descriptive message, so `git log -- sources.js` shows the full audit trail.
+
+### Changed
+
+- 📊 **Catalog size 231 → 222 models** (net **-9**). Free-tier shrinkage on OpenRouter (23 → 15) was partly offset by additions on `opencode-zen`, `googleai`, `qwen`, `ollama-cloud`, `llm7`, `scaleway`.
+- 🔧 **Provider-level fixes**:
+  - `nvidia` — `nvidia/mistral-nemotron` → `mistralai/mistral-nemotron` (namespace change)
+  - `cloudflare` — `@cf/deepseek/...` → `@cf/deepseek-ai/...`, `@cf/ibm/...` → `@cf/ibm-granite/...` (namespace changes); Gemma 4 26B ctx `128k` → `256k`
+  - `ollama-cloud` — 9 ctx corrections: `glm-5.2`, `minimax-m3`, `deepseek-v4-flash`, `deepseek-v4-pro` upgraded to `1M`; `gemma4:31b` downgraded to `128k`; `kimi-k2.6`, `kimi-k2.7-code` to `256k`; `glm-5.1`, `minimax-m2.5` to `198k`
+  - `scaleway` — 4 ctx corrections (Dedicated-tier vs Serverless-tier): `Mistral Medium 3.5` `256k` → `180k` (Serverless), `Llama 3.3 70B` `100k` → `128k` (Dedicated), `Devstral 2` `200k` → `260k` (Dedicated), `GLM 5.2` `256k` → `1M`
+  - `sambanova` — standardize 5 entries to `131k` (exact API value `131072`); `MiniMax-M2.7` `192k` → `196k`
+  - `cerebras` — 3 ctx corrections to actual free-tier limits: `zai-glm-4.7` `128k` → `64k`, `gpt-oss-120b` `128k` → `65k`, `gemma-4-31b` `128k` → `65k`
+  - `zai` — `zai/glm-4.7-flash` `203k` → `200k`
+  - `codestral` — `codestral-2508` `256k` → `128k`
+  - `novita` — `tencent/hy3` `262k` → `256k`
+  - `ovhcloud` — `Qwen3-Embedding-8B` ctx `'-'` → `'32k'`
+- 📈 **README updated to reflect new model count** — badges, prose, /stats sample shape, and benchmark copy now say `~222 models` instead of `~191`.
+
+### Removed
+
+- 🗑️ **31 dead models removed** from the live catalog (kept in git history for posterity):
+  - `nvidia` (14): `MiniMax M2.7`, `Llama 4 Maverick`, `Mistral Small 4`, `Ministral 14B`, `Seed OSS 36B`, `Stockmark 100B`, `Qwen2.5 Coder 32B`, `Phi 4 Mini`, `Mistral Large 675B`, `DeepSeek V3.2`, `DeepSeek R1`, `Mistral Nemotron` (old ns), `Nemotron 3 Nano` (old id), `Llama 4 Scout`
+  - `openrouter` (11): `Poolside Laguna XS.2`, `Tencent Hy3`, `Qwen3 Coder`, `Llama 3.3 70B Instruct`, `Hermes 3 Llama 3.1 405B`, `Qwen3 Next 80B A3B`, `OpenRouter Free`, `Llama 3.2 3B Instruct`, `Dolphin Mistral 24B Venice`, `LFM 2.5 1.2B Instruct`, `LFM 2.5 1.2B Thinking`
+  - `ovhcloud` (3): `Qwen3 Coder 30B MoE`, `Qwen3 32B`, `Mistral 7B Instruct v0.3`
+  - `opencode-zen` (1): `Tencent Hy3 Free`
+  - `novita` (1): `Qwen3.5 Plus`
+  - `routeway` (1): `Poolside Laguna M.1` (model `available: false`)
+
+### Skipped (flagged for review)
+
+- ⏸️ **`llm7` subagent reported `minimax-m2.6` removed** — but our catalog has `minimax-m2.7`, not `m2.6`. The subagent's "removal" doesn't match any entry in `sources.js`, so it was discarded. `m2.7` stays in the catalog until directly verified.
+- ⏸️ **`routeway` 2 ID "fixes" flagged as paid** — `deepseek-v4-flash:free` and `step-3.5-flash:free` are listed as free in our catalog, but the live API shows they have non-zero prices. Skipped for this release; vava will decide whether to keep them in the free catalog or move them out.
+
+### Maintenance
+
+- 🧪 **779 tests pass** (`pnpm test`). One test updated (`test/test.js:989`) to reflect the new NVIDIA catalog head — `minimaxai/minimax-m2.7` is gone, so the test now asserts the next 5 models in the discovery-ordered list.
+- 🛡️ `pnpm start` runs without runtime error. TUI / Web Dashboard / daemon all load cleanly with the new 222-model catalog.
+- 📦 `audit_state.json` refreshed — every provider's `lastAudited` is now `2026-07-27` and each has a fresh fingerprint for the next audit cycle.
+
+### Files
+
+- **Modified**: `sources.js` (30 net lines: -31 removed, +21 added, ~30 fixed inline), `audit_state.json` (timestamps + fingerprints refreshed), `README.md` (6 references to `191` → `222`), `test/test.js` (1 test updated), `package.json` (version bump `0.5.63` → `0.5.64`).
+- **New**: `changelog/v0.5.64.md`, `audit_state.json` (refreshed), `.pi/skills/update_models/SKILL.md` (v2 skill), 20 `/tmp/audit_diffs/<provider>.json` (transient, gitignored).
+
+### Breaking changes
+
+None. All fixes are in-place context-window or namespace corrections. Removed models are still cited in `git log -- sources.js` and the v0.5.64 changelog above for migration reference.

package/changelog/v0.5.65.md

@@ -0,0 +1,35 @@
+# Changelog v0.5.65 - 2026-07-29
+
+### Fixed
+
+- 🛡️ **#146 — OpenRouter / rate-limited providers no longer burned the user's daily quota.** Health-probe pings are now throttled at two orthogonal levels so a single overloaded model can no longer wipe out an OpenRouter (or similar) daily request limit. Issue submitted by `@ia-S-on`.
+
+  - **Per-provider quota circuit-breaker** — New module `src/core/provider-cooldown.js` exposes an in-memory pause map. When a probe returns HTTP `429`, FCM parses the `Retry-After` header (and falls back to the OpenRouter-style body `"Rate limit exceeded, please try again N seconds later."`), then pauses **every model of that provider** until the window expires. Probe pressure drops from ~30 req/min/provider to ~0 req/min/provider for the entire duration. Affected code: `src/core/ping.js` (trigger on `resp.status === 429`), `src/tui/app.js` (skip paused providers in `runPingCycle`).
+  - **Per-model broken cooldown with exponential backoff** — `src/core/probe-cache.js` now caches `broken` entries too. `isCacheFresh()` and `getModelsDueForProbe()` honour a new `brokenCooldownMs(failureCount)` ladder (30 s → 1 m → 2 m → 5 m, then plateau at 5 min). `recordProbeResults()` tracks `consecutiveFailures` (increments on `broken`, resets to `0` on `ok`). This caps worst-case probe pressure on a permanently-broken model at **~0.2 req/min** instead of the previous **~30 req/min**.
+  - **Live footer chip** — When one or more providers are paused, the TUI footer now shows e.g. `⏸ openrouter 14h · ⏸ ovhcloud 34s` (warm amber background, live countdown that auto-rolls to s/m/h/d). New state field `state.pausedProviders` (defaults to `{}`) + render-table chip fed by `pausedProviders` option. Behaviour verified on first run — OVHcloud was correctly paused for 34 s on a real 429 during smoke testing.
+
+### Added
+
+- 📖 **README health-check warning** — New section `⚠️ Health checks consume provider quota` between the providers table and the Tier scale. Explains the 1k/day OpenRouter cap, links the v0.5.65 fix points, and offers the immediate workaround (remove the provider's key from Settings — anonymous probes still work for liveness checks).
+- 🧪 **18 new unit tests** in `test/test.js` covering both layers (`Issue #146: provider quota circuit-breaker` + `Issue #146: per-model broken-cooldown backoff`). Three pre-existing tests in `test/probe-cache.test.js` updated to reflect the new `broken`-is-fresh-during-cooldown contract.
+
+### Changed
+
+- 📖 `pnpm test` suite: **797 / 797 passing** (was 779 / 779 in 0.5.64; net +18 new tests + 3 updated).
+- 📖 `node --check` clean on every touched file; TUI smoke-tested in tmux (no runtime regression).
+- 📖 Footer's existing probe-cache chip is unchanged in behaviour but gains the sibling paused-providers chip immediately to its right.
+
+### Migration / workaround
+
+> **If you were already hitting the 429 from issue #146 before upgrading**, simply upgrading to v0.5.65 fixes the runaway probe loop — no manual step required. The provider will be auto-paused on its next probe.
+>
+> **If you want zero authenticated probes for a particular provider** (e.g. you only use OpenRouter for ad-hoc calls and want FCM to leave the key alone), leave the key empty in Settings — anonymous probes still tell us whether each model is up and will not burn your personal daily quota.
+
+### Files
+
+- **New**: `src/core/provider-cooldown.js` (≈220 lines).
+- **Modified**: `src/core/ping.js` (import + 429 trigger), `src/core/probe-cache.js` (constants, `brokenCooldownMs`, `isCacheFresh`, `getModelsDueForProbe`, `recordProbeResults`), `src/tui/app.js` (import + `runPingCycle` skip + paused snapshot), `src/tui/tui-state.js` (`pausedProviders: {}` on state), `src/tui/render-table.js` (new chip + thread option through), `README.md` (warning section), `test/test.js` (18 new tests), `test/probe-cache.test.js` (3 updated), `package.json` (`0.5.64` → `0.5.65`), this changelog.
+
+### Breaking changes
+
+None. All changes are additive and behaviour-preserving for users who were not already rate-limited.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "free-coding-models",
-  "version": "0.5.63",
+  "version": "0.5.65",
   "description": "Find the fastest coding LLM models in seconds \u2014 ping free models from multiple providers, pick the best one for OpenCode, Cursor, or any AI coding assistant.",
   "keywords": [
     "nvidia",

package/sources.js

@@ -41,8 +41,8 @@
 // 📖 NIM source - https://build.nvidia.com
 export const nvidiaNim = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
-  ['minimaxai/minimax-m2.7', 'MiniMax M2.7', 'S+', '78.0%', '200k'],
   ['z-ai/glm-5.2', 'GLM 5.1', 'S+', '82.8%', '128k'],
+  // Removed (2026-07-27): minimaxai/minimax-m2.7 (MiniMax M2.7) — EOL 2026-07-27 (HTTP 410 Gone)
   ['moonshotai/kimi-k2.6', 'Kimi K2.6', 'S+', '80.2%', '262k'],
   ['deepseek-ai/deepseek-v4-pro', 'DeepSeek V4 Pro', 'S+', '80.6%', '1M'],
   ['deepseek-ai/deepseek-v4-flash', 'DeepSeek V4 Flash', 'S+', '79.0%', '1M'],
@@ -50,34 +50,39 @@ export const nvidiaNim = [
   ['nvidia/nemotron-3-ultra-550b-a55b', 'Nemotron 3 Ultra', 'S+', '71.9%', '1M'],
   // ── S tier — SWE-bench Verified 60–70% ──
   ['openai/gpt-oss-120b', 'GPT OSS 120B', 'S', '62.4%', '128k'],
-  ['meta/llama-4-maverick-17b-128e-instruct', 'Llama 4 Maverick', 'S+', '74.8%', '1M'],
+  // Removed (2026-07-27): meta/llama-4-maverick-17b-128e-instruct (Llama 4 Maverick) — EOL 2026-07-27 (HTTP 410 Gone)
   ['mistralai/mistral-medium-3.5-128b', 'Mistral Medium 3.5', 'S+', '77.6%', '256k'],
-  ['mistralai/mistral-small-4-119b-2603', 'Mistral Small 4', 'S', '60.0%', '256k'],
+  // Removed (2026-07-27): mistralai/mistral-small-4-119b-2603 (Mistral Small 4) — EOL 2026-07-27 (HTTP 410 Gone)
   ['minimaxai/minimax-m3', 'MiniMax M3', 'S+', '78.4%', '1M'],
-  ['nvidia/mistral-nemotron', 'Mistral Nemotron', 'S', '-', '128k'],
-  ['deepseek-ai/deepseek-v3.2', 'DeepSeek V3.2', 'S+', '70.0%', '160k'],
+  ['mistralai/mistral-nemotron', 'Mistral Nemotron', 'S', '-', '128k'], // Fixed ID (2026-07-27): nvidia/mistral-nemotron → mistralai/mistral-nemotron
+  // Removed (2026-07-27): deepseek-ai/deepseek-v3.2 (DeepSeek V3.2) — HTTP 404
   // ── A+ tier — SWE-bench Verified 50–60% ──
-  ['mistralai/mistral-large-3-675b-instruct-2512', 'Mistral Large 675B', 'A+', '58.0%', '256k'],
+  // Removed (2026-07-27): mistralai/mistral-large-3-675b-instruct-2512 (Mistral Large 675B) — EOL 2026-07-23 (HTTP 410 Gone)
   ['nvidia/nemotron-3-super-120b-a12b', 'Nemotron 3 Super', 'S', '60.5%', '128k'],
   ['nvidia/nemotron-3-nano-omni-30b-a3b-reasoning', 'Nemotron 3 Omni', 'A+', '52.0%', '256k'],
-  ['meta-llama/llama-4-scout-17b-16e-instruct', 'Llama 4 Scout', 'B', '28.0%', '10M'],
+  // Removed (2026-07-27): meta-llama/llama-4-scout-17b-16e-instruct (Llama 4 Scout) — HTTP 404
   ['nvidia/llama-3.3-nemotron-super-49b-v1.5', 'Llama 3.3 Nemotron Super 49B v1.5', 'A+', '-', '128k'],
   // ── A tier — SWE-bench Verified 40–50% ──
   ['nvidia/nemotron-3-nano-30b-a3b', 'Nemotron Nano 30B', 'A-', '38.8%', '1M'],
   ['openai/gpt-oss-20b', 'GPT OSS 20B', 'A+', '50.3%', '128k'],
   ['google/gemma-4-31b-it', 'Gemma 4 31B', 'A+', '52.0%', '256k'],
-  ['qwen/qwen2.5-coder-32b-instruct', 'Qwen2.5 Coder 32B', 'A', '47.0%', '128k'],
-  ['deepseek-ai/deepseek-r1', 'DeepSeek R1', 'A', '49.2%', '128k'],
-  ['nvidia/nemotron-3-nano', 'Nemotron 3 Nano', 'A-', '38.8%', '256k'],
+  // Removed (2026-07-27): qwen/qwen2.5-coder-32b-instruct (Qwen2.5 Coder 32B) — EOL 2026-05-12 (HTTP 410 Gone)
+  // Removed (2026-07-27): deepseek-ai/deepseek-r1 (DeepSeek R1) — HTTP 404
+  // Removed (2026-07-27): nvidia/nemotron-3-nano (Nemotron 3 Nano) — HTTP 404 (replaced by nvidia/nvidia-nemotron-nano-9b-v2)
+  ['nvidia/nvidia-nemotron-nano-9b-v2', 'Nemotron Nano 9B v2', 'A-', '-', '128k'], // Added (2026-07-27)
+  ['deepseek-ai/deepseek-coder-6.7b-instruct', 'DeepSeek Coder 6.7B', 'A-', '-', '128k'], // Added (2026-07-27)
+  ['mistralai/codestral-22b-instruct-v0.1', 'Codestral 22B', 'A', '-', '32k'], // Added (2026-07-27)
   // ── A- tier — SWE-bench Verified 35–40% ──
-  ['bytedance/seed-oss-36b-instruct', 'Seed OSS 36B', 'A+', '56.0%', '32k'],
-  ['stockmark/stockmark-2-100b-instruct', 'Stockmark 100B', 'A-', '36.0%', '32k'],
+  // Removed (2026-07-27): bytedance/seed-oss-36b-instruct (Seed OSS 36B) — EOL 2026-07-27 (HTTP 410 Gone)
+  // Removed (2026-07-27): stockmark/stockmark-2-100b-instruct (Stockmark 100B) — EOL 2026-07-15 (HTTP 410 Gone)
   // ── B+ tier — SWE-bench Verified 30–35% ──
-  ['mistralai/ministral-14b-instruct-2512', 'Ministral 14B', 'B+', '34.0%', '32k'],
+  // Removed (2026-07-27): mistralai/ministral-14b-instruct-2512 (Ministral 14B) — EOL 2026-07-27 (HTTP 410 Gone)
+  ['thinkingmachines/inkling', 'Inkling', 'B+', '-', '32k'], // Added (2026-07-27)
   // ── B tier — SWE-bench Verified 20–30% ──
   ['meta/llama-3.2-11b-vision-instruct', 'Llama 3.2 11B Vision', 'B', '28.0%', '128k'],
+  ['nvidia/nemotron-mini-4b-instruct', 'Nemotron Mini 4B', 'B', '-', '32k'], // Added (2026-07-27)
   // ── C tier — lightweight/edge models ──
-  ['microsoft/phi-4-mini-instruct', 'Phi 4 Mini', 'C', '14.0%', '128k'],
+  // Removed (2026-07-27): microsoft/phi-4-mini-instruct (Phi 4 Mini) — EOL 2026-07-15 (HTTP 410 Gone)
 ]
 
 // 📖 Groq source - https://console.groq.com
@@ -97,11 +102,11 @@ export const groq = [
 export const cerebras = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   // ⚠️ DEPRECATED — shutdown 2026-08-17 (Cerebras official notice)
-  ['zai-glm-4.7', 'GLM 4.7', 'S+', '73.8%', '128k', null, '2026-08-17'],
+  ['zai-glm-4.7', 'GLM 4.7', 'S+', '73.8%', '64k', null, '2026-08-17'], // Fixed (2026-07-27): ctx '128k' → '64k' (free tier per official docs)
   // ── S tier — SWE-bench Verified 60–70% ──
-  ['gpt-oss-120b', 'GPT OSS 120B', 'S', '62.4%', '128k'],
+  ['gpt-oss-120b', 'GPT OSS 120B', 'S', '62.4%', '65k'], // Fixed (2026-07-27): ctx '128k' → '65k' (free tier per official docs)
   // ── A tier — SWE-bench Verified 40–50% ──
-  ['gemma-4-31b', 'Gemma 4 31B', 'A+', '52.0%', '128k'],
+  ['gemma-4-31b', 'Gemma 4 31B', 'A+', '52.0%', '65k'], // Fixed (2026-07-27): ctx '128k' → '65k' (free tier per official docs)
 ]
 
 // 📖 SambaNova source - https://cloud.sambanova.ai
@@ -109,15 +114,15 @@ export const cerebras = [
 // 📖 Keep this catalog conservative: only models surfaced in current SambaNova docs.
 export const sambanova = [
   // ── S+ tier ──
-  ['MiniMax-M2.7',                         'MiniMax M2.7',       'S+', '78.0%', '192k'],
+  ['MiniMax-M2.7',                         'MiniMax M2.7',       'S+', '78.0%', '196k'], // Fixed (2026-07-27): ctx '192k' → '196k' (API exact 196608)
   // ── S tier ──
-  ['DeepSeek-V3.1',                        'DeepSeek V3.1',      'S',  '66.0%', '128k'],
+  ['DeepSeek-V3.1',                        'DeepSeek V3.1',      'S',  '66.0%', '131k'], // Fixed (2026-07-27): ctx '128k' → '131k' (API exact 131072)
   ['DeepSeek-V3.2',                        'DeepSeek V3.2',      'S+', '70.0%', '32k'],
-  ['gpt-oss-120b',                         'GPT OSS 120B',       'S',  '62.4%', '128k'],
+  ['gpt-oss-120b',                         'GPT OSS 120B',       'S',  '62.4%', '131k'], // Fixed (2026-07-27): ctx '128k' → '131k'
   // ── A tier ──
-  ['gemma-4-31B-it',                       'Gemma 4 31B',        'A+',  '52.0%', '128k'],
+  ['gemma-4-31B-it',                       'Gemma 4 31B',        'A+',  '52.0%', '131k'], // Fixed (2026-07-27): ctx '128k' → '131k'
   // ── A- tier ──
-  ['Meta-Llama-3.3-70B-Instruct',          'Llama 3.3 70B',      'B', '22.0%', '128k'],
+  ['Meta-Llama-3.3-70B-Instruct',          'Llama 3.3 70B',      'B', '22.0%', '131k'], // Fixed (2026-07-27): ctx '128k' → '131k'
   // ── B+ tier ──
 ]
 
@@ -135,33 +140,36 @@ export const openrouter = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   ['nvidia/nemotron-3-ultra-550b-a55b:free', 'Nemotron 3 Ultra', 'S+', '71.9%', '1M'],
   ['poolside/laguna-m.1:free', 'Poolside Laguna M.1', 'S+', '72.5%', '262k'],
-  ['poolside/laguna-xs.2:free', 'Poolside Laguna XS.2', 'S', '68.2%', '262k'],
   ['poolside/laguna-xs-2.1:free', 'Poolside Laguna XS 2.1', 'S+', '70.9%', '262k'],
+  // Removed (2026-07-27): poolside/laguna-xs.2:free (Poolside Laguna XS.2) — superseded by poolside/laguna-xs-2.1:free
   // ── S tier — SWE-bench Verified 60–70% ──
   ['cohere/north-mini-code:free', 'North Mini Code', 'S', '-', '256k'],
-  ['tencent/hy3:free', 'Tencent Hy3', 'S', '-', '262k'],
-  ['qwen/qwen3-coder:free', 'Qwen3 Coder', 'S', '69.6%', '1M'],
+  // Removed (2026-07-27): tencent/hy3:free (Tencent Hy3) — no longer on OpenRouter free tier
+  // Removed (2026-07-27): qwen/qwen3-coder:free (Qwen3 Coder) — no longer on OpenRouter free tier
+  ['poolside/laguna-s-2.1:free', 'Poolside Laguna S 2.1', 'S+', '-', '262k'], // Added (2026-07-27)
   // ── A+ tier — SWE-bench Verified 50–60% ──
   ['nvidia/nemotron-3-super-120b-a12b:free', 'Nemotron 3 Super', 'S', '60.5%', '1M'],
   ['nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free', 'Nemotron 3 Omni', 'A+', '52.0%', '256k'],
+  ['inclusionai/ling-3.0-flash:free', 'Ling-3.0 Flash', 'A+', '-', '262k'], // Added (2026-07-27)
   // ── A tier — SWE-bench Verified 40–50% ──
   ['nvidia/nemotron-3-nano-30b-a3b:free', 'Nemotron Nano 30B', 'A-', '38.8%', '256k'],
   ['nvidia/nemotron-nano-12b-v2-vl:free', 'Nemotron Nano 12B VL', 'A', '20.0%', '128k'],
   ['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'],
-  ['meta-llama/llama-3.3-70b-instruct:free', 'Llama 3.3 70B Instruct', 'B', '22.0%', '131k'],
-  ['nousresearch/hermes-3-llama-3.1-405b:free', 'Hermes 3 Llama 3.1 405B', 'A', '-', '131k'],
-  ['qwen/qwen3-next-80b-a3b-instruct:free', 'Qwen3 Next 80B A3B', 'S+', '70.6%', '262k'],
+  // Removed (2026-07-27): meta-llama/llama-3.3-70b-instruct:free (Llama 3.3 70B Instruct) — no longer on free tier
+  // Removed (2026-07-27): nousresearch/hermes-3-llama-3.1-405b:free (Hermes 3 Llama 3.1 405B) — no longer on free tier
+  // Removed (2026-07-27): qwen/qwen3-next-80b-a3b-instruct:free (Qwen3 Next 80B A3B) — no longer on free tier
   // ── B+ tier — SWE-bench Verified 30–35% ──
   ['nvidia/nemotron-nano-9b-v2:free', 'Nemotron Nano 9B', 'B+', '18.0%', '128k'],
+  ['openai/gpt-oss-20b:free', 'GPT OSS 20B', 'A-', '-', '131k'], // Added (2026-07-27)
   // ── B tier — SWE-bench Verified 20–30% ──
-  ['openrouter/free', 'OpenRouter Free', 'B', '-', '200k'],
-  ['meta-llama/llama-3.2-3b-instruct:free', 'Llama 3.2 3B Instruct', 'B', '-', '131k'],
-  ['cognitivecomputations/dolphin-mistral-24b-venice-edition:free', 'Dolphin Mistral 24B Venice', 'B', '-', '32k'],
+  // Removed (2026-07-27): openrouter/free (OpenRouter Free) — no longer on free tier
+  // Removed (2026-07-27): meta-llama/llama-3.2-3b-instruct:free (Llama 3.2 3B Instruct) — no longer on free tier
+  // Removed (2026-07-27): cognitivecomputations/dolphin-mistral-24b-venice-edition:free (Dolphin Mistral 24B Venice) — no longer on free tier
   // ── C tier — lightweight/edge models ──
   ['nvidia/nemotron-3.5-content-safety:free', 'Nemotron 3.5 Content Safety', 'C', '-', '128k'],
-  ['liquid/lfm-2.5-1.2b-instruct:free', 'LFM 2.5 1.2B Instruct', 'C', '-', '32k'],
-  ['liquid/lfm-2.5-1.2b-thinking:free', 'LFM 2.5 1.2B Thinking', 'C', '-', '32k'],
+  // Removed (2026-07-27): liquid/lfm-2.5-1.2b-instruct:free (LFM 2.5 1.2B Instruct) — no longer on free tier
+  // Removed (2026-07-27): liquid/lfm-2.5-1.2b-thinking:free (LFM 2.5 1.2B Thinking) — no longer on free tier
 ]
 
 // 📖 GitHub Models source - https://models.github.ai
@@ -234,7 +242,7 @@ export const mistral = [
 // 📖 API keys now use the Mistral platform key format; CODESTRAL_API_KEY remains supported as an alias.
 export const codestral = [
   // ── A tier — SWE-bench Verified 40–50% ──
-  ['codestral-2508', 'Codestral', 'A', '40.0%', '256k'],
+  ['codestral-2508', 'Codestral', 'A', '40.0%', '128k'], // Fixed (2026-07-27): ctx '256k' → '128k' per official Mistral model card
   ['codestral-2', 'Codestral 2', 'B+', '-', '128k'],
 ]
 
@@ -242,21 +250,22 @@ export const codestral = [
 // 📖 1M free tokens — API keys at https://console.scaleway.com/iam/api-keys
 export const scaleway = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
-  ['devstral-2-123b-instruct-2512', 'Devstral 2 123B', 'S+', '72.2%', '200k'],
+  ['devstral-2-123b-instruct-2512', 'Devstral 2 123B', 'S+', '72.2%', '260k'], // Fixed (2026-07-27): ctx '200k' → '260k' (Dedicated tier)
   ['qwen3-235b-a22b-instruct-2507', 'Qwen3 235B', 'A', '45.2%', '250k'],
-  ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '256k'],
+  ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '1M'], // Fixed (2026-07-27): ctx '256k' → '1M' (preview extended)
   // ── S tier — SWE-bench Verified 60–70% ──
   ['qwen3.5-397b-a17b', 'Qwen3.5 400B VLM', 'S+', '76.2%', '250k'],
   ['gpt-oss-120b', 'GPT OSS 120B', 'S', '62.4%', '128k'],
-  ['mistral-medium-3.5-128b', 'Mistral Medium 3.5 128B', 'S+', '77.6%', '256k'],
+  ['mistral-medium-3.5-128b', 'Mistral Medium 3.5 128B', 'S+', '77.6%', '180k'], // Fixed (2026-07-27): ctx '256k' → '180k' (Serverless tier)
   // ── A+ tier — SWE-bench Verified 50–60% ──
   ['mistral-large-3-675b-instruct-2512', 'Mistral Large 675B', 'A+', '58.0%', '250k'],
   ['qwen3-coder-30b-a3b-instruct', 'Qwen3 Coder 30B', 'A+', '51.6%', '128k'],
   ['qwen3.6-35b-a3b', 'Qwen3.6 35B MoE', 'S+', '73.4%', '256k'],
   ['holo2-30b-a3b', 'Holo2 30B', 'A+', '52.0%', '22k'],
   ['gemma-4-26b-a4b-it', 'Gemma 4 26B MoE', 'A+', '-', '256k'],
+  ['gemma-4-31b-it', 'Gemma 4 31B IT', 'A+', '-', '128k'], // Added (2026-07-27)
   // ── A- tier — SWE-bench Verified 35–40% ──
-  ['llama-3.3-70b-instruct', 'Llama 3.3 70B', 'B', '22.0%', '100k'],
+  ['llama-3.3-70b-instruct', 'Llama 3.3 70B', 'B', '22.0%', '128k'], // Fixed (2026-07-27): ctx '100k' → '128k' (Dedicated tier)
   // ── B+ tier — SWE-bench Verified 30–35% ──
   ['mistral-small-3.2-24b-instruct-2506', 'Mistral Small 3.2', 'B', '20.0%', '128k'],
   ['pixtral-12b-2409', 'Pixtral 12B', 'B+', '-', '128k'],
@@ -267,9 +276,11 @@ export const scaleway = [
 // 📖 Google AI Studio source - https://aistudio.google.com
 // 📖 OpenAI-compatible endpoint exposes Gemini models; free quotas vary by model and region.
 export const googleai = [
+  ['gemini-3.6-flash',                          'Gemini 3.6 Flash',             'S+', '-',         '1M'], // Added (2026-07-27)
   ['gemini-3.5-flash',                          'Gemini 3.5 Flash',             'S+', '78.0%',     '1M'],
   ['gemini-3.1-pro-preview',                    'Gemini 3.1 Pro Preview',       'S+', '80.6%', '1M'],
   ['gemini-3-flash-preview',                    'Gemini 3 Flash Preview',       'S+',  '78.0%', '1M'],
+  ['gemini-3.5-flash-lite',                     'Gemini 3.5 Flash Lite',        'S', '-',         '1M'], // Added (2026-07-27)
   ['gemini-3.1-flash-lite',                     'Gemini 3.1 Flash Lite',        'S', '62.8%', '1M'],
   ['gemini-2.5-pro',                            'Gemini 2.5 Pro',               'S', '63.8%', '1M'],
   ['gemini-2.5-flash',                          'Gemini 2.5 Flash',             'A+', '54.0%', '1M'],
@@ -280,7 +291,7 @@ export const googleai = [
 // 📖 Free tier is limited to Flash models; paid GLM models are intentionally excluded.
 export const zai = [
   // ── S tier — SWE-bench Verified 60–70% ──
-  ['zai/glm-4.7-flash', 'GLM-4.7-Flash', 'A+', '59.2%', '203k'],
+  ['zai/glm-4.7-flash', 'GLM-4.7-Flash', 'A+', '59.2%', '200k'], // Fixed (2026-07-27): ctx '203k' → '200k' per official docs
   ['zai/glm-4.5-flash', 'GLM-4.5-Flash', 'S', '59.2%', '128k'],
   // ── A tier — SWE-bench Verified 40–50% ──
   ['zai/glm-4.6v-flash', 'GLM-4.6V-Flash', 'A', '-', '128k'],
@@ -306,6 +317,7 @@ export const qwen = [
   ['qwen3-coder-next', 'Qwen3 Coder Next', 'S+', '70.6%', '256k'],
   ['qwen3-coder-480b-a35b-instruct', 'Qwen3 Coder 480B', 'S', '69.6%', '256k'],
   // ── A+ tier — SWE-bench Verified 50–60% ──
+  ['qwen3.7-flash', 'Qwen3.7 Flash', 'A+', '-', '1M'], // Added (2026-07-27)
   ['qwen3.6-flash', 'Qwen3.6 Flash', 'A+', '60.0%', '1M'],
   ['qwen3.5-flash', 'Qwen3.5 Flash', 'S', '64.4%', '1M'],
   ['qwen3-coder-flash', 'Qwen3 Coder Flash', 'A+', '55.0%', '1M'],
@@ -340,13 +352,13 @@ export const cloudflare = [
   ['@cf/qwen/qwen2.5-coder-32b-instruct', 'Qwen2.5 Coder 32B', 'A', '47.0%', '128k'],
   ['@cf/openai/gpt-oss-20b', 'GPT OSS 20B', 'A+', '50.3%', '128k'],
   ['@cf/qwen/qwq-32b', 'QwQ 32B', 'A', '-', '128k'],
-  ['@cf/deepseek/deepseek-r1-distill-qwen-32b', 'DeepSeek R1 Distill Qwen 32B', 'A', '-', '128k'],
+  ['@cf/deepseek-ai/deepseek-r1-distill-qwen-32b', 'DeepSeek R1 Distill Qwen 32B', 'A', '-', '128k'], // Fixed (2026-07-27): namespace 'deepseek' → 'deepseek-ai'
   // ── A- tier — SWE-bench Verified 35–40% ──
   ['@cf/meta/llama-3.3-70b-instruct-fp8-fast', 'Llama 3.3 70B', 'B', '22.0%', '128k'],
-  ['@cf/google/gemma-4-26b-a4b-it', 'Gemma 4 26B MoE', 'A-', '38.0%', '128k'],
+  ['@cf/google/gemma-4-26b-a4b-it', 'Gemma 4 26B MoE', 'A-', '38.0%', '256k'], // Fixed (2026-07-27): ctx '128k' → '256k' (April 2026 changelog)
   // ── B+ tier — SWE-bench Verified 30–35% ──
   ['@cf/mistralai/mistral-small-3.1-24b-instruct', 'Mistral Small 3.1', 'B+', '30.0%', '128k'],
-  ['@cf/ibm/granite-4.0-h-micro', 'Granite 4.0 Micro', 'B+', '30.0%', '128k'],
+  ['@cf/ibm-granite/granite-4.0-h-micro', 'Granite 4.0 Micro', 'B+', '30.0%', '128k'], // Fixed (2026-07-27): namespace 'ibm' → 'ibm-granite'
   // ── B tier — SWE-bench Verified 20–30% ──
   ['@cf/meta/llama-3.1-8b-instruct-fast', 'Llama 3.1 8B Instruct (Fast)', 'C', '18.0%', '128k'],
 ]
@@ -358,17 +370,17 @@ export const cloudflare = [
 export const ovhcloud = [
   ['Qwen3.5-397B-A17B',                         'Qwen3.5 397B MoE',    'S+',  '76.2%',     '262k'],
   ['Qwen3.6-27B',                               'Qwen3.6 27B',         'S+',  '77.2%',     '262k'],
-  ['Qwen3-Coder-30B-A3B-Instruct',             'Qwen3 Coder 30B MoE',  'A+', '51.6%', '256k'],
+  // Removed (2026-07-27): Qwen3-Coder-30B-A3B-Instruct (Qwen3 Coder 30B MoE) — no longer in catalog
   ['gpt-oss-120b',                              'GPT OSS 120B',         'S',  '62.4%', '131k'],
   ['gpt-oss-20b',                               'GPT OSS 20B',          'A+',  '50.3%', '131k'],
   ['Meta-Llama-3_3-70B-Instruct',               'Llama 3.3 70B',        'B', '22.0%', '131k'],
-  ['Qwen3-32B',                                 'Qwen3 32B',            'B+', '30.0%', '32k'],
+  // Removed (2026-07-27): Qwen3-32B (Qwen3 32B) — no longer in catalog
   ['Mistral-Small-3.2-24B-Instruct-2506',       'Mistral Small 3.2',    'B', '20.0%', '128k'],
-  ['Mistral-7B-Instruct-v0.3',                  'Mistral 7B Instruct',  'B',  '25.0%', '127k'],
+  // Removed (2026-07-27): Mistral-7B-Instruct-v0.3 (Mistral 7B Instruct) — no longer in catalog
   ['Mistral-Nemo-Instruct-2407',                'Mistral Nemo',         'B+', '30.0%', '118k'],
   ['Qwen3.5-9B',                                'Qwen3.5 9B',           'B+', '30.0%', '262k'],
   // ── Embeddings ──
-  ['Qwen3-Embedding-8B',                        'Qwen3 Embedding 8B',   'B',  '-',     '-'],
+  ['Qwen3-Embedding-8B',                        'Qwen3 Embedding 8B',   'B',  '-',     '32k'], // Fixed (2026-07-27): ctx '-' → '32k'
   ['bge-m3',                                    'BGE M3',               'B',  '-',     '-'],
   ['bge-multilingual-gemma2',                   'BGE Multilingual Gemma2','B','-',     '-'],
   // Fix (2026-05-26): Qwen3.5-9B ctx 128k→262k, Mistral-Small ctx 131k→128k, Mistral-Nemo ctx 128k→118k, Mistral-7B ctx 32k→127k
@@ -387,7 +399,9 @@ export const opencodeZen = [
   ['mimo-v2.5-free',                   'MiMo-V2.5 Free',          'S+', '-',     '200k'],
   ['nemotron-3-ultra-free',            'Nemotron 3 Ultra Free',   'S+', '71.9%', '200k'],
   ['north-mini-code-free',             'North Mini Code Free',    'B+', '-',     '200k'],
-  ['hy3-free',                         'Tencent Hy3 Free',        'S',  '-',     '200k'],
+  ['poolside/laguna-s-2.1-free',       'Laguna S 2.1 Free',       'S+', '-',     '262k'], // Added (2026-07-27)
+  ['ling-3.0-flash-free',              'Ling-3.0-flash Free',     'A+', '-',     '262k'], // Added (2026-07-27)
+  // Removed (2026-07-27): hy3-free (Tencent Hy3 Free) — no longer on OpenCode Zen
 ]
 
 // 📖 Kilo source - https://api.kilo.ai/api/gateway
@@ -405,6 +419,7 @@ export const llm7 = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   ['minimax-m2.7', 'MiniMax M2.7', 'S+', '78.0%', '180k'],
   // ── A+ tier — SWE-bench Verified 50–60% ──
+  ['gemini-3.1-flash-lite', 'Gemini 3.1 Flash Lite', 'A+', '-', '1M'], // Added (2026-07-27)
   ['gpt-oss:20b', 'GPT OSS 20B', 'A+', '50.3%', '128k'],
   // ── A tier — SWE-bench Verified 40–50% ──
   ['codestral-latest', 'Codestral Latest', 'A', '40.0%', '32k'],
@@ -417,9 +432,11 @@ export const routeway = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   ['deepseek-v4-flash:free', 'DeepSeek V4 Flash', 'S+', '79.0%', '1M'],
   ['step-3.5-flash:free', 'Step 3.5 Flash', 'S+', '74.4%', '256k'],
-  ['laguna-m.1:free', 'Poolside Laguna M.1', 'S+', '72.5%', '131k'],
+  // Removed (2026-07-27): laguna-m.1:free (Poolside Laguna M.1) — unavailable on Routeway
   ['laguna-xs.2:free', 'Poolside Laguna XS.2', 'S', '68.2%', '131k'],
+  ['step-3.7-flash:free', 'Step 3.7 Flash', 'S+', '74.4%', '256k'], // Added (2026-07-27)
   // ── S tier — SWE-bench Verified 60–70% ──
+  ['ling-3.0-flash:free', 'Ling 3.0 Flash', 'S', '-', '256k'], // Added (2026-07-27)
   ['ling-2.6-flash:free', 'Ling 2.6 Flash', 'S', '61.2%', '262k'],
   ['gpt-oss-120b:free', 'GPT OSS 120B', 'S', '60.0%', '131k'],
   // ── A tier — SWE-bench Verified 40–50% ──
@@ -440,10 +457,9 @@ 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 = [
-  // ── S+ tier — SWE-bench Verified ≥70% ──
-  ['qwen/qwen3.5-plus', 'Qwen3.5 Plus', 'S+', '80.0%', '1M'],
   // ── S tier — SWE-bench Verified 60–70% ──
-  ['tencent/hy3', 'Tencent Hy3', 'S', '-', '262k'],
+  ['tencent/hy3', 'Tencent Hy3', 'S', '-', '256k'], // Fixed (2026-07-27): ctx '262k' → '256k' per official novita description
+  // Removed (2026-07-27): qwen/qwen3.5-plus (Qwen3.5 Plus) — no longer in novita catalog
 ]
 
 // 📖 Ollama Cloud source - https://ollama.com/pricing and https://ollama.com/search?c=cloud
@@ -452,22 +468,26 @@ export const novita = [
 export const ollamaCloud = [
   // ── S+ tier — SWE-bench Verified ≥70% ──
   ['nemotron-3-ultra', 'Nemotron 3 Ultra', 'S+', '71.9%', '256k'],
-  ['glm-5.1', 'GLM 5.1', 'S+', '82.8%', '128k'],
-  ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '128k'],
+  ['glm-5.1', 'GLM 5.1', 'S+', '82.8%', '198k'], // Fixed (2026-07-27): ctx '128k' → '198k'
+  ['glm-5.2', 'GLM 5.2', 'S+', '82.8%', '1M'], // Fixed (2026-07-27): ctx '128k' → '1M'
   ['minimax-m2.7', 'MiniMax M2.7', 'S+', '78.0%', '200k'],
-  ['minimax-m3', 'MiniMax M3', 'S+', '78.4%', '512k'],
-  ['minimax-m2.5', 'MiniMax M2.5', 'S+', '80.2%', '200k'],
-  ['kimi-k2.6', 'Kimi K2.6', 'S+', '80.2%', '262k'],
-  ['deepseek-v4-flash', 'DeepSeek V4 Flash', 'S+', '79.0%', '256k'],
-  ['deepseek-v4-pro', 'DeepSeek V4 Pro', 'S+', '80.6%', '256k'],
+  ['minimax-m3', 'MiniMax M3', 'S+', '78.4%', '1M'], // Fixed (2026-07-27): ctx '512k' → '1M'
+  ['minimax-m2.5', 'MiniMax M2.5', 'S+', '80.2%', '198k'], // Fixed (2026-07-27): ctx '200k' → '198k'
+  ['kimi-k2.6', 'Kimi K2.6', 'S+', '80.2%', '256k'], // Fixed (2026-07-27): ctx '262k' → '256k'
+  ['deepseek-v4-flash', 'DeepSeek V4 Flash', 'S+', '79.0%', '1M'], // Fixed (2026-07-27): ctx '256k' → '1M'
+  ['deepseek-v4-pro', 'DeepSeek V4 Pro', 'S+', '80.6%', '1M'], // Fixed (2026-07-27): ctx '256k' → '1M'
   // ── S tier — SWE-bench Verified 60–70% ──
-  ['kimi-k2.7-code', 'Kimi K2.7 Code', 'S', '60.4%', '262k'],
+  ['kimi-k2.7-code', 'Kimi K2.7 Code', 'S', '60.4%', '256k'], // Fixed (2026-07-27): ctx '262k' → '256k'
   ['gpt-oss:120b', 'GPT OSS 120B', 'S', '62.4%', '128k'],
   ['nemotron-3-super', 'Nemotron 3 Super', 'S', '60.5%', '256k'],
+  ['kimi-k3', 'Kimi K3', 'S+', '-', '1M'], // Added (2026-07-27)
+  ['gemini-3-flash-preview', 'Gemini 3 Flash Preview', 'S+', '-', '1M'], // Added (2026-07-27)
   // ── A+ tier — SWE-bench Verified 50–60% ──
   ['kimi-k2.5', 'Kimi K2.5', 'A+', '-', '200k'],
-  ['gemma4:31b', 'Gemma 4 31B', 'A+', '52.0%', '256k'],
+  ['gemma4:31b', 'Gemma 4 31B', 'A+', '52.0%', '128k'], // Fixed (2026-07-27): ctx '256k' → '128k'
   ['gpt-oss:20b', 'GPT OSS 20B', 'A+', '50.3%', '128k'],
+  ['mistral-large-3:675b-cloud', 'Mistral Large 3 675B Cloud', 'A+', '-', '256k'], // Added (2026-07-27)
+  ['qwen3.5', 'Qwen 3.5 Cloud', 'A+', '-', '256k'], // Added (2026-07-27)
   ['nemotron-3-nano:30b', 'Nemotron 3 Nano 30B', 'A-', '38.8%', '1M'],
 ]
 

package/src/core/ping.js

@@ -13,6 +13,9 @@
  *   - Quota extraction from rate limit headers (multiple variants supported)
  *   - Cached provider quota polling with TTL and error backoff
  *   - Cloudflare account ID resolution from environment
+ *   - Per-provider circuit-breaker on 429 (issue #146): pauses ALL models of a provider
+ *     when the provider returns a quota-exhausted response, so the ping loop stops
+ *     hammering the user's daily quota while the provider's retry window is active.
  *
  *   → Functions:
  *   - `resolveCloudflareUrl`: Resolve {account_id} placeholder from CLOUDFLARE_ACCOUNT_ID env var
@@ -31,6 +34,7 @@
  *   - ../src/constants.js: PING_TIMEOUT
  *   - ../src/provider-quota-fetchers.js: _fetchProviderQuotaFromModule (quota fetching with cache)
  *   - ../src/quota-capabilities.js: supportsUsagePercent
+ *   - ./provider-cooldown.js: pauseProviderQuota, extractRetryAfterFromResponse (issue #146)
  *
  *   ⚙️ Configuration:
  *   - PING_TIMEOUT: Timeout in ms for ping requests (default: 15000)
@@ -43,6 +47,10 @@
 import { PING_TIMEOUT } from './constants.js'
 import { fetchProviderQuota as _fetchProviderQuotaFromModule, extractQuota as _extractQuotaFromModule, processResponseHeaders as _processResponseHeadersFromModule } from './provider-quota-fetchers.js'
 import { supportsUsagePercent } from './quota-capabilities.js'
+import {
+  extractRetryAfterFromResponse,
+  pauseProviderQuota,
+} from './provider-cooldown.js'
 
 const DISABLED_THINKING_RETRY_STATUSES = new Set([400, 422])
 const disabledThinkingUnsupportedProviders = new Set()
@@ -171,6 +179,14 @@ export async function ping(apiKey, modelId, providerKey, url) {
       req = buildPingRequest(apiKey, modelId, providerKey, url, { disableThinking: false })
       resp = await sendPingFetch(req, ctrl.signal)
     }
+    // 📖 Provider circuit-breaker (issue #146): openrouter + similar gateways rate-limit
+    // 📖 at the provider/key level, not per-model. A 429 means the WHOLE key is paused,
+    // 📖 often for hours — re-pinging every model every 2-30s would burn the quota in minutes.
+    // 📖 Honour Retry-After header + the OpenRouter body "try again N seconds later".
+    if (resp.status === 429) {
+      const retryMs = await extractRetryAfterFromResponse(resp)
+      if (retryMs > 0) pauseProviderQuota(providerKey, retryMs)
+    }
     // 📖 Normalize all HTTP 2xx statuses to "200" so existing verdict/avg logic still works.
     const code = resp.status >= 200 && resp.status < 300 ? '200' : String(resp.status)
     // 📖 Passive quota tracker (t2): parse the response headers via the shared module.