发布

  • chore(catalog): V21 — prune 7 dead models, re-enable Cerebras zai-glm-4.7 (#184)

    frostbyte_neo 发布于 2026-06-03 09:35:58 +00:00

    Live-probed 2026-06-03 with production keys. Delete (confirmed dead):

    • llm7: gpt-oss-20b, meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo,
      ministral-8b-2512 — LLM7 silently serves mistral-small-3.2 for these
      (200 OK but the wrong model — a routing footgun)
    • llm7: GLM-4.6V-Flash — now pro-gated (402 "upgrade required")
    • openrouter: arcee-ai/trinity-large-thinking:free, minimax/minimax-m2.5:free,
      baidu/cobuddy:free — 404 "no endpoints found" (delisted / moved to paid)

    Re-enable cerebras/zai-glm-4.7 (V9 disabled it; live-probed 200 free again).

    Deletes remove the fallback_config row first (foreign_keys=ON). V12 idempotency
    test updated to expect the pruned OpenRouter rows absent. Models 115 -> 108.

    下载附件