fix(gateway): extract vertex-anthropic streaming token usage

[serhiizghama]

Streaming usage for vertex-anthropic was silently coming back empty. extractTokenUsage switches on the provider, and the Anthropic block is case "anthropic": only — so vertex-anthropic fell through to the OpenAI-shaped default, which reads data.usage.prompt_tokens/completion_tokens. Vertex Anthropic streams the Anthropic Messages wire format (message.usage.input_tokens, output_tokens, cache and thinking fields), so the default branch found nothing and returned all-null usage.

The effect is real cost drift: with usage null the streaming path falls back to estimateTokens, and cachedTokens, cacheCreationTokens and reasoning/thinking tokens are dropped entirely, so the cache-read discount never applies. The direct anthropic provider bills the same chunk correctly; vertex-anthropic did not.

Every sibling handler already groups the two the same way — map-finish-reason-to-openai, extract-tool-calls, extract-reasoning, the finish-reason switch in chat.ts, and the non-streaming parse-provider-response — so this was just a missing case label on the streaming usage extractor. Added it, and added a spec that asserts vertex-anthropic extracts the streaming message_start shape and matches anthropic for an identical chunk.

Summary by CodeRabbit

• Bug Fixes
  • Improved token usage reporting for Vertex Anthropic conversations.
  • Prompt, cached, completion, reasoning, and total token counts are now extracted consistently without double-counting reasoning tokens.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 38ef9bd5-1b62-4432-8c26-6b51fbd58daf

📥 Commits

Reviewing files that changed from the base of the PR and between 07d9b2c and 6859e67.

📒 Files selected for processing (2)

• apps/gateway/src/chat/tools/extract-token-usage.spec.ts
• apps/gateway/src/chat/tools/extract-token-usage.ts

---

Walkthrough

Adds vertex-anthropic support to token usage extraction by reusing Anthropic handling and covering prompt, cache, completion, reasoning, total tokens, and provider parity in tests.

Changes

Vertex Anthropic token usage

Layer / File(s)	Summary
Provider routing and extraction validation apps/gateway/src/chat/tools/extract-token-usage.ts, apps/gateway/src/chat/tools/extract-token-usage.spec.ts	vertex-anthropic uses the Anthropic token extraction path, with tests covering token fields and matching Anthropic extraction results.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested reviewers: smakosh

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely summarizes the main change: fixing streaming token usage extraction for vertex-anthropic.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}