ProxMenux/AppImage/config/verified_ai_models.json

{
  "_description": "Verified AI models for ProxMenux notifications. Only models listed here will be shown to users. Models are tested to work with the chat/completions API format.",
  "_updated": "2026-09-02",
  "_verifier": "Refreshed with tools/ai-models-verifier (private). Re-run before each ProxMenux release to keep the list current. The verifier and ProxMenux share the same reasoning/thinking-model handlers so their verdicts stay aligned with runtime behaviour.",
  "groq": {
    "models": [
      "allam-2-7b",
      "qwen/qwen3.8-27b",
      "openai/gpt-oss-20b",
      "groq/compound-mini"
    ],
    "recommended": "openai/gpt-oss-20b",
    "_note": "Verified 2026-09-02 with the Groq API. 4 of 12 tested pass. `llama-3.3-70b-versatile` no longer served by Groq. Recommended kept as `openai/gpt-oss-20b` (english-capable) — `allam-2-7b` is fastest but Arabic-focused; add manually if serving Arabic notifications."
  },
  "gemini": {
    "models": [
      "gemini-2.5-flash-lite",
      "gemini-3.1-flash-lite",
      "gemini-3.5-flash-lite",
      "gemini-3.1-flash-lite-preview",
      "gemini-flash-lite-latest",
      "gemini-2.5-flash",
      "gemini-3-flash-preview",
      "gemini-3.5-flash"
    ],
    "recommended": "gemini-2.5-flash-lite",
    "_deprecated": [
      "gemini-2.0-flash",
      "gemini-2.0-flash-lite",
      "gemini-1.5-flash",
      "gemini-1.0-pro",
      "gemini-pro"
    ],
    "_note": "Verified 2026-09-02. gemini-flash-lite-latest passes consistently but gemini-2.5-flash-lite remains recommended because 'latest' aliases can drift. gemini-3.1-flash-lite is the stable successor to 3-flash-preview. Pro variants continue to reject thinkingBudget=0 and are overkill for notification translation."
  },
  "openai": {
    "models": [
      "gpt-4.1-nano-2025-04-14",
      "gpt-4.1-nano",
      "gpt-4.1-2025-04-14",
      "gpt-4.1",
      "gpt-4o-2024-11-20",
      "gpt-4o-2024-05-13",
      "gpt-4o-mini",
      "gpt-3.5-turbo-16k",
      "gpt-4o-mini-2024-07-18",
      "gpt-5-nano",
      "gpt-3.5-turbo-1106",
      "gpt-3.5-turbo-0125",
      "gpt-5-nano-2025-08-07",
      "gpt-4.1-mini",
      "gpt-4.1-mini-2025-04-14",
      "gpt-3.5-turbo",
      "gpt-4o",
      "gpt-4o-2024-08-06",
      "gpt-4",
      "gpt-4-turbo",
      "gpt-4-turbo-2024-04-09",
      "gpt-4-0613"
    ],
    "recommended": "gpt-4.1-nano",
    "_note": "Verified 2026-09-02 with the OpenAI API — 22 of 74 tested pass. Includes stable aliases (gpt-4.1-nano, gpt-4o-mini, ...), dated snapshots for pinning (gpt-4o-2024-11-20, gpt-4.1-nano-2025-04-14) and legacy families (gpt-3.5-turbo, gpt-4) for cost-optimised use cases. Recommended kept as `gpt-4.1-nano` (stable alias, 1.39s)."
  },
  "anthropic": {
    "models": [
      "claude-haiku-4-5-20251001",
      "claude-opus-4-8",
      "claude-sonnet-4-5-20250929",
      "claude-opus-4-5-20251101",
      "claude-opus-4-7",
      "claude-sonnet-4-6",
      "claude-fable-5",
      "claude-fable-5-1",
      "claude-opus-4-6"
    ],
    "recommended": "claude-haiku-4-5-20251001",
    "_note": "Verified 2026-09-02 — 9 of 11 tested pass. The `claude-haiku-4-5` alias no longer resolves upstream; the working snapshot is `claude-haiku-4-5-20251001` — used as recommended (2.89s). Full family surfaced: haiku/sonnet/opus/fable across generations 4.5/4.6/4.7/4.8/5."
  },
  "openrouter": {
    "models": [
      "google/gemini-2.5-flash-lite",
      "google/gemini-3.5-flash-lite",
      "amazon/nova-pro-v1",
      "mistralai/mistral-medium-3",
      "google/gemini-2.5-flash",
      "amazon/nova-lite-v1",
      "openai/gpt-5.4-mini",
      "openai/gpt-4o-mini",
      "openai/gpt-4.1",
      "perplexity/sonar",
      "meta-llama/llama-4-scout",
      "x-ai/grok-4.20",
      "google/gemini-3-flash-preview",
      "meta-llama/llama-4-maverick",
      "perplexity/sonar-pro",
      "openai/gpt-4.1-nano",
      "anthropic/claude-haiku-4.5",
      "mistralai/mistral-large-2512",
      "meta-llama/llama-3.1-70b-instruct",
      "qwen/qwen3-235b-a22b-2507",
      "moonshotai/kimi-k2",
      "anthropic/claude-opus-4.8",
      "meta-llama/llama-3.3-70b-instruct",
      "openai/gpt-5.1",
      "anthropic/claude-sonnet-5",
      "mistralai/mistral-small-3.2-24b-instruct",
      "cohere/command-a",
      "x-ai/grok-4.3",
      "anthropic/claude-sonnet-4.6",
      "deepseek/deepseek-v3.2",
      "openai/gpt-oss-120b",
      "meta-llama/llama-3.1-8b-instruct",
      "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free",
      "nvidia/nemotron-3-super-120b-a12b:free",
      "google/gemma-4-26b-a4b-it:free",
      "nvidia/nemotron-nano-12b-v2-vl:free",
      "nvidia/nemotron-3-nano-30b-a3b:free",
      "poolside/laguna-s-2.1:free",
      "openai/gpt-oss-20b:free"
    ],
    "recommended": "meta-llama/llama-3.3-70b-instruct",
    "_note": "Curated manually from the 249+ models OpenRouter serves — the auto-verifier passes ~200 today but only a subset are practical for notification translation. Includes the 11 previously-curated paid models plus 22 new mainstream additions (gpt-5.1, claude-opus-4.8, sonnet-5, grok-4.3/4.20, kimi-k2, deepseek-v3.2, nova, cohere-a, sonar, ...) and 6 `:free` variants re-added manually — the free tier is intentionally supported per user request even though today's rate limits blocked some of them from passing the test."
  },
  "ollama": {
    "_note": "Ollama models are local, we don't filter them. User manages their own models.",
    "models": [],
    "recommended": ""
  }
}