fix: normalize Together provider for pricing lookup

[dani29]

Summary

• normalize the runtime custom-provider name together to the pricing catalog's canonical together_ai identifier
• apply provider normalization consistently when resolving catalog pricing, while preserving the runtime provider name for routing, metrics, and scoped overrides
• add an end-to-end cost regression using real GLM-5.2 token counts and cached-input pricing

Root cause

Pricing rows are normalized when the hosted datasheet is ingested, but the runtime provider identifier was used verbatim during cost lookup. A custom OpenAI-compatible provider named together therefore could not match rows stored under together_ai; cost calculation returned zero even when the catalog contained the model.

The hosted catalog still needs a direct Together GLM-5.2 row. That data request is tracked in #6036. Once present, this change lets existing together deployments consume it without renaming a DB-backed provider.

Validation

• go test ./modelcatalog/datasheet -run 'TestCalculateCost_TogetherAliasUsesTogetherAIPricingWithCache|TestGetPricing_' -count=1
• go test ./modelcatalog/datasheet -count=1
• the regression calculates $13.2067812 from 36,807,862 input tokens, including 33,829,760 cached tokens, and 54,932 output tokens

[coderabbitai]

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes
  • Fixed pricing resolution for Together provider requests by recognizing the together alias.
  • Together usage now applies the correct cached-read, input, and output token rates.
• Tests
  • Added regression coverage to verify end-to-end pricing for the Together provider alias.

Walkthrough

The pricing resolver now maps the together provider alias to together_ai for catalog lookups. Existing raw-provider handling remains unchanged. A regression test verifies cached-read, input, and output pricing.

Changes

Together provider alias pricing

Layer / File(s)	Summary
Provider normalization framework/modelcatalog/datasheet/types.go	normalizeProvider maps together to together_ai before existing normalization cases.
Normalized pricing resolution and regression coverage framework/modelcatalog/datasheet/cost.go, framework/modelcatalog/datasheet/cost_test.go	resolvePricing uses the normalized provider for base pricing lookup. The regression test verifies Together AI cached-read, input, and output rates for the alias.

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

Suggested reviewers: tejasghatte, pratham-mishra04, impoiler

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description explains the root cause, implementation, regression test, and validation, but omits several required template sections.	Add the Changes, Type of change, Affected areas, How to test, Breaking changes, Related issues, Security considerations, and Checklist sections.

✅ Passed checks (4 passed)

Check name	Status	Explanation
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.
Title check	✅ Passed	The title clearly and concisely describes the provider normalization fix for Together pricing lookup.

✨ 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.}