Account for Gemini reasoning tokens in usage details

[PratikDhanave]

When converting Gemini usage metadata, toUsageDetails mapped four fields but dropped ThoughtsTokenCount entirely:

return message.UsageDetails{
    InputTokenCount:       int64(u.PromptTokenCount),
    OutputTokenCount:      int64(u.CandidatesTokenCount),
    TotalTokenCount:       int64(u.TotalTokenCount),
    CachedInputTokenCount: int64(u.CachedContentTokenCount),
}

Per the genai docs, TotalTokenCount is "the sum of prompt_token_count, candidates_token_count, tool_use_prompt_token_count, and thoughts_token_count" — so thoughtsTokenCount is a separate bucket, not included in candidatesTokenCount. For a thinking model, those reasoning tokens were silently lost: InputTokenCount + OutputTokenCount + ReasoningTokenCount no longer reconciled with TotalTokenCount, and the framework’s dedicated UsageDetails.ReasoningTokenCount field (summed by UsageDetails.Add) stayed 0.

Both OpenAI providers already populate ReasoningTokenCount (chat.go, responses.go); Gemini was the outlier.

Fix

    CachedInputTokenCount: int64(u.CachedContentTokenCount),
    ReasoningTokenCount:   int64(u.ThoughtsTokenCount),

Test

TestUsageContent_ReasoningTokens returns a response with thoughtsTokenCount: 8 (prompt 10, candidates 5, total 23) and asserts ReasoningTokenCount == 8. It fails on the old code (0) and passes with the fix; the record now reconciles (10 + 5 + 8 = 23).

[Copilot]

Pull request overview

This pull request updates the Gemini provider’s usage metadata conversion to correctly preserve Gemini “reasoning” (thoughts) tokens in the framework’s UsageDetails, aligning Gemini with the existing OpenAI providers’ behavior.

Changes:

• Map Gemini ThoughtsTokenCount into message.UsageDetails.ReasoningTokenCount in toUsageDetails.
• Add a regression test ensuring thoughtsTokenCount is surfaced and reconciles with totalTokenCount.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
provider/geminiprovider/agent.go	Populate UsageDetails.ReasoningTokenCount from Gemini usage metadata (ThoughtsTokenCount).
provider/geminiprovider/agent_test.go	Add test coverage verifying reasoning tokens are not dropped for Gemini responses.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Parity review complete - no issues found. See full analysis below.

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• awmgmcpg

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "awmgmcpg"

See Network Configuration for more information.

Generated by Go API Consistency Review Agent · 37.2 AIC · ⌖ 4.82 AIC · ⊞ 5.7K · ◷