Skip to content

v3.31.0 — model catalog aligned to gateway (new frontier models + free-tier fix)

Choose a tag to compare

View all tags
@VickyXAI VickyXAI released this 11 Jul 23:45
v3.31.0
6a44e13

Re-aligns Franklin's static model tables with the live BlockRun gateway (18 new models) and fixes a live breakage where the free default pointed at an EOL'd model.

  • 18 new models wired end-to-end (pricing, context windows, max-output, vision, /model picker, shortcuts): OpenAI GPT-5.6 (sol/terra/luna) + GPT-5.4 mini/nano; Anthropic Fable 5, Sonnet 5, Opus 4.5, Sonnet 4.5; Google Gemini 3.5 Flash + 3.1 Flash Lite; 7 new NVIDIA free models. Flagship shortcuts follow the newest tier: `gpt`→GPT-5.6 Sol, `sonnet`/`claude`→Sonnet 5, `fable`→Fable 5.
  • Free default was dead — fixed. `nvidia/deepseek-v4-flash` was EOL'd (410). Replaced with `nvidia/qwen3-next-80b-a3b-instruct`, chosen by live testing as the cleanest free instruction-follower (no thinking-leak / markdown fences; the old maverick default was flaky and leaked reasoning into structured outputs). qwen3-next leads every free chain + the clean-JSON utility roles; maverick stays as a diverse-family secondary.
  • Free tier never falls back to paid — removed `zai/glm-5.2` from the empty-response recovery chain; free chains are $0-only (guarded by test).
  • Killed proxy shortcut drift — the proxy now imports the picker's single source of truth instead of a drifted duplicate.
  • Adaptive-thinking guard — Sonnet 5 + Fable 5 excluded from the explicit thinking flag (that generation 400s on it).

Verified: local suite 513/513, tsc clean, live gateway calls to pick the free default.

Assets 2
Loading