fix(catalog): align openrouter prices with live openrouter.ai rates

[astraltrekkin]

TLDR

Problem this solves:

• 41 OpenRouter catalog prices did not match openrouter.ai
• Spend, budgets, and alerts used those stale per-token rates
• 20 of those rows still linked the page that contradicted them

How it solves it:

• Rewrote the drifted input, output, and cache-read rates
• Matched GET https://openrouter.ai/api/v1/models headline pricing
• Left max tokens, vision flags, and the auto-updater alone

User Flow

Before: a developer who bills OpenRouter traffic through LiteLLM records the wrong spend because the catalog still has last quarter's rates

1. They send POST https://litellm-domain/v1/chat/completions with "model": "openrouter/qwen/qwen3.5-plus-02-15"
2. The completion succeeds and usage is 1M prompt + 1M completion tokens
3. They open https://litellm-domain/ui/?page=logs and see that request billed at $2.80 ($0.40 + $2.40 per 1M)
4. The same check for openrouter/openai/gpt-oss-120b shows $0.98 ($0.18 + $0.80 per 1M)

After: the same traffic is billed at the rates OpenRouter publishes today

1. They send the same POST https://litellm-domain/v1/chat/completions with "model": "openrouter/qwen/qwen3.5-plus-02-15"
2. The completion succeeds with the same 1M + 1M usage
3. https://litellm-domain/ui/?page=logs now shows that request at $1.82 ($0.26 + $1.56 per 1M)
4. openrouter/openai/gpt-oss-120b now shows $0.207 ($0.037 + $0.17 per 1M)

Relevant issues

#39501

Linear ticket

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Shared setup: curl -s https://openrouter.ai/api/v1/models -o or.json, then compare every overlapping openrouter/* catalog row's input_cost_per_token / output_cost_per_token / cache_read_input_token_cost to the API pricing.prompt / pricing.completion / pricing.input_cache_read

Before ()

1. Captured at merge base 27274f65e40fa38870b9e1450ab885e09074644d
2. GET https://openrouter.ai/api/v1/models returned HTTP 200 with 424 models
3. Compare against model_prices_and_context_window.json at that commit: comparable=87 mismatch_fields=84 mismatch_entries=41
4. openrouter/qwen/qwen3.5-plus-02-15 registry $0.40 / $2.40 vs live $0.26 / $1.56 (page https://openrouter.ai/qwen/qwen3.5-plus-02-15 agrees with live)
5. openrouter/openai/gpt-oss-120b registry $0.18 / $0.80 vs live $0.037 / $0.17 (page https://openrouter.ai/openai/gpt-oss-120b agrees with live)
6. Control openrouter/anthropic/claude-opus-5 already matched live $5 / $25

After ()

1. Captured at fe8a7846391127cb071b06c9a1f30551e13283b2
2. Same live GET https://openrouter.ai/api/v1/models (HTTP 200)
3. Compare against the updated catalog: comparable=87 mismatch_fields=4 mismatch_entries=2
4. openrouter/qwen/qwen3.5-plus-02-15 registry $0.26 / $1.56 matches live
5. openrouter/openai/gpt-oss-120b registry $0.037 / $0.17 matches live
6. Control openrouter/anthropic/claude-opus-5 still $5 / $25
7. The two leftover rows are later OpenRouter moves on openrouter/deepseek/deepseek-v4-pro and openrouter/deepseek/deepseek-v4-pro-0813, not the original stale values

Type

🐛 Bug Fix

Caveats (if any)

Medium

• OpenRouter can reprice at any time, so this is a headline snapshot from the live models API
• Two DeepSeek rows moved again after this patch and still disagree today
• Tiered *_above_256k_tokens fields were not part of the reported drift and were left alone

Low

• The weekly auto-updater was not turned on. It still maps max_tokens to context length and would fail the schema on openrouter/auto

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing rainbowgore:cursor/berriai-litellm-39501-3141 (80a63f0) with litellm_internal_staging (4b1e24e)}

[greptile-apps]

Greptile Summary

This PR updates matching OpenRouter input, output, and cache-read prices in both model catalogs and adds a local regression test for selected rates

• Keeps the root and bundled backup catalogs byte-aligned
• Updates 41 OpenRouter model entries
• Adds value checks for 10 changed models and one unchanged control

Confidence Score: 4/5

The pricing changes appear safe to merge, with non-blocking test-coverage and comment-policy cleanup recommended

Both catalogs remain identical and no incorrect changed price was established, but most updated entries lack value-level regression assertions

Files Needing Attention: tests/test_litellm/litellm_core_utils/test_get_model_cost_map.py

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Updates OpenRouter token and cache-read rates while preserving the existing model metadata and catalog schema
litellm/model_prices_and_context_window_backup.json	Mirrors the root catalog changes exactly for bundled fallback loading
tests/test_litellm/litellm_core_utils/test_get_model_cost_map.py	Adds local root and backup price assertions, but covers only 10 of the 41 changed entries and adds a disallowed explanatory comment

_{Reviews (1): Last reviewed commit: "fix(catalog): align openrouter prices wi..." | Re-trigger Greptile}

[greptile-apps]

Price fixture covers ten models

The regression fixture pins only 10 of 41 changed models, so incorrect rates in omitted entries can pass while producing wrong spend

Context Used: CLAUDE.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[greptile-apps]

Fixture comment duplicates context

This prose restates the fixture's source and purpose, violating the source-comment policy and adding text that can drift from the test

Context Used: CLAUDE.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[devin-ai-integration]

Superseded by rolling registry PR #39388, which applies the OpenRouter price fixes re-derived from the live catalog API plus the regression test