Repository navigation
fix: override finishReason to tool-calls for Gemini 3 thoughtSignature - #288
Merged
Merged
Conversation
When Gemini 3 makes tool calls with encrypted reasoning (thoughtSignature), it returns finish_reason='stop' instead of 'tool_calls'. This causes the AI SDK to not continue the conversation after tool execution. This fix detects when there are: - Tool calls present - Encrypted reasoning details (thoughtSignature) - finish_reason is 'stop' And overrides the finish reason to 'tool-calls' so the conversation continues. Applied to both doGenerate (non-streaming) and doStream (streaming) paths.
Contributor
There was a problem hiding this comment.
Pull request overview
This PR fixes a critical issue with Gemini 3 models that use thinking mode (thoughtSignature/encrypted reasoning). When these models return tool calls alongside encrypted reasoning, they incorrectly signal finish_reason: "stop" instead of finish_reason: "tool-calls", causing the AI SDK to prematurely end the conversation instead of continuing after tool execution.
Key Changes:
- Added finishReason override logic in
doGenerate(non-streaming path) to detect the problematic pattern (tool calls + encrypted reasoning + stop) and override to 'tool-calls' - Added equivalent finishReason override logic in
doStreamflush handler (streaming path) to ensure consistent behavior - Both implementations use the same detection criteria: presence of tool calls, presence of encrypted reasoning with data, and finish_reason='stop'
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+429
to
+443
| // Fix for Gemini 3 thoughtSignature: when there are tool calls with encrypted | ||
| // reasoning (thoughtSignature), the model returns 'stop' but expects continuation. | ||
| // Override to 'tool-calls' so the SDK knows to continue the conversation. | ||
| const hasToolCalls = choice.message.tool_calls && choice.message.tool_calls.length > 0; | ||
| const hasEncryptedReasoning = reasoningDetails.some( | ||
| (d) => d.type === ReasoningDetailType.Encrypted && d.data | ||
| ); | ||
| const shouldOverrideFinishReason = | ||
| hasToolCalls && | ||
| hasEncryptedReasoning && | ||
| choice.finish_reason === 'stop'; | ||
|
|
||
| const effectiveFinishReason = shouldOverrideFinishReason | ||
| ? 'tool-calls' | ||
| : mapOpenRouterFinishReason(choice.finish_reason); |
There was a problem hiding this comment.
The new finishReason override logic for Gemini 3 thoughtSignature lacks unit test coverage. While there are existing tests for tool_calls and reasoning_details separately, there's no test that verifies the specific scenario being fixed: when a model returns tool_calls with encrypted reasoning and finish_reason='stop', the SDK should override it to 'tool-calls'.
Consider adding a test in src/chat/index.test.ts similar to the existing 'should extract reasoning content from reasoning_details' test, but with:
tool_callspresent in the responsereasoning_detailswithtype: ReasoningDetailType.Encryptedanddatapresentfinish_reason: 'stop'in the response- Assertion that the returned
finishReasonis'tool-calls'
Collaborator
There was a problem hiding this comment.
agree, this needs tests
Comment on lines
+971
to
+980
| // Fix for Gemini 3 thoughtSignature: when there are tool calls with encrypted | ||
| // reasoning (thoughtSignature), the model returns 'stop' but expects continuation. | ||
| // Override to 'tool-calls' so the SDK knows to continue the conversation. | ||
| const hasToolCalls = toolCalls.length > 0; | ||
| const hasEncryptedReasoning = accumulatedReasoningDetails.some( | ||
| (d) => d.type === ReasoningDetailType.Encrypted && d.data | ||
| ); | ||
| if (hasToolCalls && hasEncryptedReasoning && finishReason === 'stop') { | ||
| finishReason = 'tool-calls'; | ||
| } |
There was a problem hiding this comment.
The streaming finishReason override logic for Gemini 3 thoughtSignature lacks unit test coverage. Consider adding a test in src/chat/index.test.ts in the doStream section, similar to the existing 'should stream tool deltas' test, but with:
- Tool call chunks in the stream
- A chunk with
reasoning_detailscontainingtype: ReasoningDetailType.Encryptedanddata finish_reason: 'stop'in the final chunk- Assertion that the finish event has
finishReason: 'tool-calls'
When Gemini 3 makes tool calls with encrypted reasoning (thoughtSignature), it returns finish_reason='stop' instead of 'tool_calls'. This causes the AI SDK to not continue the conversation after tool execution. This fix detects when there are: - Tool calls present - Encrypted reasoning details (thoughtSignature) - finish_reason is 'stop' And overrides the finish reason to 'tool-calls' so the conversation continues. Applied to both doGenerate (non-streaming) and doStream (streaming) paths.
mattapperson
approved these changes
Dec 9, 2025
Merged
kesavan-byte
pushed a commit
to osm-API/ai-sdk-provider
that referenced
this pull request
Feb 13, 2026
OpenRouterTeam#288) * fix: override finishReason to tool-calls for Gemini 3 thoughtSignature When Gemini 3 makes tool calls with encrypted reasoning (thoughtSignature), it returns finish_reason='stop' instead of 'tool_calls'. This causes the AI SDK to not continue the conversation after tool execution. This fix detects when there are: - Tool calls present - Encrypted reasoning details (thoughtSignature) - finish_reason is 'stop' And overrides the finish reason to 'tool-calls' so the conversation continues. Applied to both doGenerate (non-streaming) and doStream (streaming) paths. * fix: override finishReason to tool-calls for Gemini 3 thoughtSignature When Gemini 3 makes tool calls with encrypted reasoning (thoughtSignature), it returns finish_reason='stop' instead of 'tool_calls'. This causes the AI SDK to not continue the conversation after tool execution. This fix detects when there are: - Tool calls present - Encrypted reasoning details (thoughtSignature) - finish_reason is 'stop' And overrides the finish reason to 'tool-calls' so the conversation continues. Applied to both doGenerate (non-streaming) and doStream (streaming) paths.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
finish_reason='stop'but has tool calls and encrypted reasoning, override totool-callsso the SDK continuesProblem
Gemini 3 with thinking enabled returns
finish_reason: "stop"when making tool calls with encrypted reasoning (thoughtSignature). This causes AI SDK consumers to not continue the conversation after tool execution.Solution
Detect the pattern (tool calls + encrypted reasoning + stop) and override finishReason to
tool-callsin both:doGenerate(non-streaming)doStream(streaming) flush handler