Skip to main content
2026/10/15 16:53 (UTC+8) · Docs Update · Google 📚 New FAQ How Do I Fix Gemini Thought Signature 400 Errors in Tool Calls? separates the two errors For multi-turn function calling on Gemini 3 and later models (gemini-4-argon, gemini-3.8-flash, and others), the thoughtSignature in the previous turn’s functionCall part must be returned unchanged. This is a hard rule from the model provider:
  • Function call is missing a thought_signature: the request carries no signature because the client dropped the field when assembling history. Routes differ in leniency, so the problem may surface only on a strictly validating model. Fix the replay logic rather than switching models
  • Invalid / Corrupted thought signature: a signature was sent but the current route does not accept it, typically after the conversation switched route, model, key, or group midway. Resend once as is; if it still fails, strip signatures from the history turns and resend, or start a new conversation
On the OpenAI-compatible /v1/chat/completions endpoint the gateway manages signatures; standard tool_calls / tool messages are enough. The Function Calling guide and the Argon page now carry both fixes in their signature sections.
← Back to Live Updates · 📚 Monthly Archive