You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Claude Fable 5.2 is in the model menu: the most capable Claude, and cheaper than Fable 5.1.
Claude Haiku 5.5 is in the model menu: the fastest and cheapest Claude, now reading up to 1M tokens (Haiku 4.5 read 200K).
Changes
Older versions are replaced. Fable 5.1 and Haiku 4.5 leave the model menu. If you'd picked one of them, as your model or your fallback model, Gryphon switches you to Fable 5.2 or Haiku 5.5 the next time it loads. The default model is still Claude Sonnet 5.5.
"Test key" for an Anthropic API key now checks the key with Haiku 5.5.
For apps that install Gryphon's npm packages: the packages no longer list undici. Gryphon stopped using it, and the version they named had security advisories, including a certificate-check bypass. The plugin you install in Obsidian never included it. Gryphon's development tools are also updated to clear new advisories.
Fixes
Antigravity works again. The Antigravity command-line tool updated itself (to 1.5.9) and stopped accepting the model Gryphon asked for, Gemini 3.7 Flash, and the way Gryphon passed the effort level. Every Antigravity request failed, including in Gryphon 2.12.6. Gryphon now uses Gemini 3.8 Flash with Antigravity, at the effort level you pick, and moves you to it automatically if you'd chosen Gemini 3.7 Flash.
Codex: older model choices no longer fail. The Codex command-line tool no longer offers GPT-5.5, GPT-5.4 or GPT-5.4 mini. If you'd pinned one of them, Gryphon now uses GPT-5.6 Terra instead of failing.
Structured (JSON) answers with an Anthropic API key work again on Sonnet 5.5, Opus 5.5 and Fable 5.2, where Anthropic had started refusing the request. If the model answers in plain text instead, Gryphon asks it once more before reporting an error.