Function call is missing a thought_signature model Gemini 3.1 Pro

[mayeco]

Plugin Type

JetBrains Plugin

App Version

5.8.0

Description

Date/time: 2026-02-19T23:17:35.072Z
Extension version: 5.8.0
Provider: gemini
Model: gemini-3.1-pro-preview

Gemini generate context stream error: {"error":{"message":"{\n "error": {\n "code": 400,\n "message": "Function call is missing a thought_signature in functionCall parts. This is required for tools to work correctly, and missing thought_signature may lead to degraded model performance. Additional data, function call default_api:read_file , position 2. Please refer to https://ai.google.dev/gemini-api/docs/thought-signatures for more details.",\n "status": "INVALID_ARGUMENT"\n }\n}\n","code":400,"status":"Bad Request"}}

Reproduction steps

• any prompt to gemini 3.1-pro

Provider

Gemini

Model

gemini-3.1-pro-preview

System Information

MacOS

[PoCk3T]

Same problem in the VS Code extension

[newcommit]

Same issue

[sulimma]

This also happens for gemini-3.1-pro-preview-customtools

[peterkc]

Diagnostic Analysis

I investigated the full message pipeline to understand where thoughtSignature is lost:

Pipeline: processor.ts (capture) → SQLite (storage) → message-v2.ts:toModelMessages() (replay) → AI SDK convertToModelMessages → @ai-sdk/google converter → Gemini API

Findings

1. Unit tests: toModelMessages() correctly preserves providerMetadata (containing thoughtSignature) for same-model Gemini sessions. The differentModel guard at line 600 evaluates false when models match, and metadata flows through.

2. E2E test: Ran a live tool-call against gemini-3.1-pro-preview using @ai-sdk/google@2.0.54. The SDK emits thoughtSignature in providerMetadata on every tool-call event, and the response message includes it in providerOptions.google.thoughtSignature. Full round-trip works.

3. Root cause (likely resolved): Commit 021e42c0b added the differentModel guard to strip cross-provider metadata. The original comparison used model.api.id on one side but msg.info.modelID (set from model.id) on the other — a mismatch that could cause differentModel to evaluate true for same-model Gemini sessions, stripping the thoughtSignature metadata.

This appears fixed in v5.9.0

The JetBrains plugin v5.9.0 release notes include:

fix: preserve extra_content for Gemini 3 thought_signature support

The issue was filed against v5.8.0. Could you update to v5.9.0 and verify whether the issue persists?

If it does, I have diagnostic tests and an E2E tracer ready to dig deeper. I've also opened #6368 to add regression test coverage for Gemini thoughtSignature metadata preservation — this scenario had zero test coverage before.