All releases
v0.3.5

Fable 5, GLM-5, and the Latest Qwen

Trinity's model picker gains Anthropic's most capable model yet — Claude Fable 5 — plus Z.ai's GLM-5 generation and Alibaba's Qwen 3.7 and 3.6, so every tier can run on the newest model from your provider of choice.

New

  • Claude Fable 5 — Anthropic's most capable model is now selectable for the reasoning tier, with a 1M-token context window, alongside the existing Opus, Sonnet, and Haiku models. Opus 4.8 stays the default.
  • Z.ai's GLM-5 generation — Connect a Z.ai key and pick GLM-5.1 (the flagship agentic coder), GLM-5, or the faster GLM-5 Turbo.
  • Qwen 3.7 and 3.6 — Alibaba's Qwen3.7 Max and Plus (frontier, 1M context) and the Qwen3.6 Plus and Flash defaults join the picker, along with a local Qwen3.6 (27B) you can run through Ollama.

Improved

  • The model picker matches each provider's current lineup — The models you can choose now reflect what each vendor actually serves today. Moonshot's original Kimi K2 variants (K2 Thinking and K2 Turbo) were retired upstream, so they're no longer listed — leaving the current Kimi K2.6 and K2.5.