# Changelog v0.3.69 - 2026-05-26

### Changed

- **Full provider audit** — Every model across all 16 providers was verified against official APIs and documentation on 2026-05-26. Outdated, deprecated, and phantom models were removed; new models were added; context windows and scores were corrected.
- **~170 → ~157 models** — Net reduction of ~13 models after removing deprecated/phantom entries and adding new ones. The catalog is now leaner and more accurate.

### Removed (deprecated / shutdown / phantom)

- **Cerebras**: `qwen-3-235b-a22b-instruct-2507`, `llama3.1-8b` — deprecated May 27, 2026
- **SambaNova**: `MiniMax-M2.5` — deprecated May 18, replaced by M2.7
- **OpenRouter**: `tencent/hy3-preview:free`, `inclusionai/ling-2.6-1t:free`, `google/gemma-3n-e2b-it:free`, `google/gemma-3-27b-it:free`, `google/gemma-3-12b-it:free`, `google/gemma-3n-e4b-it:free`, `google/gemma-3-4b-it:free` — free tiers removed by providers; `google/gemma-4-31b-a4b-it:free` — never existed (data error)
- **Mistral**: `devstral-medium-latest`, `devstral-small-latest`, `magistral-small-latest` — deprecated, replaced by unified Devstral 2 and Mistral Small 4
- **Google AI / Gemini CLI**: `gemini-3.1-flash-lite-preview` — endpoint shutdown May 25, 2026
- **Cloudflare**: `@cf/google/gemma-4-31b-it` — never existed on Cloudflare; `@cf/meta/llama-3.1-8b-instruct` — deprecated May 30 + actual context only 8k
- **OVHcloud**: `Llama-3.1-8B-Instruct` — delisted from catalog
- **OpenCode Zen**: `hy3-preview-free`, `ling-2.6-flash-free`, `trinity-mini-free`, `trinity-large-preview-free` — deleted from Zen platform

### Added (new models)

- **SambaNova**: `MiniMax-M2.7` (S+, 192k), `gemma-3-12b-it` (B+, 128k)
- **OpenRouter**: `deepseek/deepseek-v4-flash:free` (S, 1M), `arcee-ai/trinity-large-thinking:free` (A, 262k), `baidu/cobuddy:free` (B+, 131k)
- **Google AI / Gemini CLI**: `gemini-3.5-flash` (S+, 1M — new GA model), `gemini-3.1-flash-lite` (A+, 1M — graduated from preview)
- **Qwen (DashScope)**: `qwen3.7-max` (S+, 1M — new flagship), `qwen3.6-plus` (S+, 1M), `qwen3.6-flash` (A+, 1M)
- **Cloudflare**: `@cf/deepseek-ai/deepseek-r1-distill-qwen-32b` (A-, 80k)
- **OVHcloud**: `Qwen3.5-397B-A17B` (S, 262k — new on OVHcloud)
- **OpenCode Zen**: `deepseek-v4-flash-free` (S+, 200k), `qwen3.6-plus-free` (S+, 1M)

### Fixed

- **Mistral model IDs** updated from `-latest` aliases to versioned IDs (`mistral-large-2512`, `mistral-medium-3-5`, `mistral-small-2603`, `devstral-2512`, `magistral-medium-2509`)
- **Codestral** ID updated to `codestral-2508`, context corrected from 256k → 128k
- **Groq**: 4 models context corrected from 128k → 131k (actual Groq limit)
- **Cerebras**: 2 models context corrected (gpt-oss-120b 128k→131k, zai-glm-4.7 200k→131k)
- **SambaNova**: Llama-4-Maverick context corrected from 1M → 128k (SambaNova actual limit)
- **Scaleway**: 7 models context windows corrected (major fixes: holo2 131k→22k, gemma-3-27b 128k→40k, devstral 256k→200k, qwen3-coder 32k→128k)
- **OpenRouter**: 7 context corrections, MiniMax M2.5 score updated 74.0% → 80.2%, Laguna models 256k→131k, Owl Alpha 128k→1M
- **Qwen**: 4 context corrections (qwen3-max 1M→256k, qwen3-coder-plus 256k→1M, qwen3-coder-flash 256k→1M, qwen3-32b 128k→256k), model ID fixed `qwen3-235b-a22b-instruct` → `qwen3-235b-a22b`
- **Cloudflare**: kimi-k2.6 context corrected 256k → 262k
- **OVHcloud**: 4 context corrections (Qwen3.5-9B 128k→262k, Mistral-Nemo 128k→118k, Mistral-7B 32k→127k, Mistral-Small 131k→128k)
- **OpenCode Zen**: nematron-3-super-free context corrected from 1M → 200k (Zen limits it)
- **GitHub Models**: ministral-3b context corrected from 32k → 128k

### Docs

- Added 14 provider audit reports in `provider_updates/` (one per provider)
- Updated README.md model counts and provider table