feat: add /tokenize enpoint for GCP and AWS Anthropic backends for chat completion schema

[hustxiayang]

Description
Adds /tokenize support for gcp and aws anthropic schema for chat completion schema, which is also split from #1650 as planned.

Provider-specific model name handling:

The modelNameOverride configured on a backend is shared across all API types (messages, chat, count_tokens). However, the count_tokens endpoints on GCP and AWS have stricter model name requirements than their inference counterparts:

• GCP Vertex AI: The count-tokens endpoint does not accept the @default or @latest version aliases (e.g., claude-sonnet-4-6@default returns "not supported for token counting"). Explicit version tags like @20250514 are accepted, as are bare model names. Since @default is required for inference endpoints that share the same backend config, the count_tokens translator strips @default and @latest suffixes only.

  • Ref: https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/count-tokens

• AWS Bedrock: The CountTokens API (POST /model/{modelId}/count-tokens) does not support cross-region inference (CRIS) model IDs (e.g., us.anthropic.claude-sonnet-4-6 returns "The provided model doesn't support counting tokens"). CRIS prefixes are required for inference endpoints (InvokeModel, Converse) that share the same backend config, so the count_tokens translator strips short region prefixes (e.g., us., eu.) only.

  • Ref: https://docs.aws.amazon.com/bedrock/latest/userguide/count-tokens.html

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 84.46970% with 41 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.83%. Comparing base (934174a) to head (1e4f0b6).

Files with missing lines	Patch %	Lines
internal/translator/tokenize_awsanthropic.go	81.13%	10 Missing and 10 partials ⚠️
internal/translator/tokenize_gcpanthropic.go	85.71%	7 Missing and 7 partials ⚠️
internal/translator/anthropic_helper.go	87.50%	3 Missing and 4 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2390      +/-   ##
==========================================
- Coverage   84.84%   84.83%   -0.02%     
==========================================
  Files         151      153       +2     
  Lines       22090    22330     +240     
==========================================
+ Hits        18743    18944     +201     
- Misses       2215     2234      +19     
- Partials     1132     1152      +20     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[aabchoo]

In the future, could you please split your PRs for each backend? It would be a lot easier to review individually.

[aabchoo]

@gemini review

[aabchoo]

/retest

[gemini-code-assist]

Code Review

This pull request adds support for translating OpenAI-compatible tokenize requests to GCP Anthropic (Claude on Vertex AI) and AWS Bedrock (Anthropic) token counting APIs. It introduces the ToGCPAnthropicV1Tokenize and ToAWSAnthropicV1Tokenize translators, updates the endpoint specification and tests, and documents the new capabilities. The feedback suggests a minor improvement in tokenize_gcpanthropic.go to use strings.TrimSuffix instead of manual slicing for stripping model version suffixes, which would make the code cleaner and less error-prone.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[aabchoo]

overall looks good to me!

[hustxiayang]

"In the future, could you please split your PRs for each backend? It would be a lot easier to review individually." sure, thanks a lot for bring it up! These two are shared because both providers (gcp, bedrock) are using the same underlying claude backend, api schemas (gcpanthropic and awsanthropic) should be identical mostly.

[hustxiayang]

/gemini review

[gemini-code-assist]

Code Review

This pull request adds support for translating OpenAI-compatible tokenize requests to GCP Anthropic (Claude on Vertex AI) and AWS Bedrock (Anthropic) token counting APIs. It introduces the respective translators, refactors tool parameter translation helpers, updates the endpoint specification logic, adds comprehensive unit tests, and updates the documentation. The review feedback suggests avoiding in-place mutation of the shared request structure in both translators to prevent side effects, explicitly accessing the prompt field instead of relying on unsafe field promotion, and adding a defensive nil check in the helper function to prevent potential nil pointer dereferences.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[aabchoo]

"In the future, could you please split your PRs for each backend? It would be a lot easier to review individually." sure, thanks a lot for bring it up! These two are shared because both providers (gcp, bedrock) are using the same underlying claude backend, api schemas (gcpanthropic and awsanthropic) should be identical mostly.

I'd still advocate for splitting it into two. They may share a lot of similarities but they're still two separate features with differences, and, as a reviewer, it's much easier to review and have a good understanding of two large PRs than one XL PR.