fix(pricing): restore Azure gpt-5.6 terra/luna rates to Azure's published prices

[olifarhaan]

TLDR

Problem this solves:

• Azure gpt-5.6 luna bills 5x more than LiteLLM reports, terra 20% more
• Every azure/ gpt-5.6 deployment under-reports spend

How it solves it:

• Restore the six azure terra/luna rows to Azure's published prices
• Pin the two azure/us/ rows in the existing regression test

Relevant issues

Fixes #36094

Linear ticket

Pre-Submission checklist

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

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• 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

There are no Azure credentials in this environment, so the live run below is the exact command list rather than captured output. Add a luna deployment to dev_config.yaml first:

  - model_name: azure-luna
    litellm_params:
      model: azure/gpt-5.6-luna
      api_base: os.environ/AZURE_API_BASE
      api_key: os.environ/AZURE_API_KEY
      api_version: "2026-05-01-preview"

1. Start the proxy: python litellm/proxy/proxy_cli.py --config litellm/proxy/dev_config.yaml --detailed_debug --reload 2>&1 | tee litellm.log
2. Send a request and read back the cost header:

curl -s -D /dev/stderr http://localhost:4000/v1/chat/completions \
  -H "Authorization: Bearer sk-1234" \
  -H "Content-Type: application/json" \
  -d '{"model": "azure-luna", "messages": [{"role": "user", "content": "Write one sentence about the sea"}]}' \
  2>&1 >/dev/null | grep -i "x-litellm-response-cost"

3. Cross-check against what Azure billed for the same token counts on the Azure OpenAI pricing page

Before this PR the reported cost is 1/5 of the Azure rate for luna. After, they match. The same holds for azure/gpt-5.6-terra at 1.25x, and for the azure/us/ and azure/eu/ data-zone variants at their 10% uplift

The spend log at http://localhost:4000/ui/?page=logs shows the same number if you prefer reading it there

Type

🐛 Bug Fix

Changes

OpenAI cut gpt-5.6 terra 20% and luna 80% on 2026-07-30. Commit 4d43080 mirrored that cut onto the azure/ rows, reasoning that "openai and bedrock_mantle entries already match". Azure did not follow. Microsoft still publishes $1.00/1M input and $6.00/1M output for luna global and $1.10/$6.60 for the us/eu data zones, and there are open Microsoft Q&A threads asking why the Azure prices were never updated (one, two). So LiteLLM has been reporting 5x less than Azure bills for luna, and 20% less for terra, on every azure/ gpt-5.6 deployment since 2026-08-01

This restores the 60 fields that commit changed, across the six global/us/eu terra and luna rows, in both cost maps. Input, output, cache-read, priority, and above-272k variants all move together. sol is untouched, and so are the openai/ and bedrock_mantle rows, whose cut is correct

Two corrections to the issue as filed, both worth a reviewer's attention. It blames PR #35258, but that PR left the azure rows alone: at its merge commit azure/gpt-5.6-luna was still 1e-06, and only the openai keys moved. The actual change was 4d43080 two days later, which was deliberate rather than collateral. The issue also reports luna only, while terra regressed identically and is included here

Because this reverses a considered decision rather than an obvious slip, it is worth a maintainer confirming the pricing call before merge. Happy to drop the terra half if you would rather keep the scope to what was reported

The parametrized cost test already pinned the global and eu rows, so it fails without this change. It had no coverage for the azure/us/ rows, which is where the report actually pointed, so those two are added

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

[greptile-apps]

Greptile Summary

The PR restores Azure GPT-5.6 terra and luna pricing across global, US, and EU entries while keeping the canonical and packaged backup maps synchronized.

• Updates base, cache-read, priority, and long-context rates.
• Expands cost-calculation regression coverage for selected priority and long-context paths.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Restores the canonical Azure GPT-5.6 terra/luna rates and preserves the documented regional uplift relationships.
litellm/model_prices_and_context_window_backup.json	Mirrors the canonical pricing changes exactly for packaged and offline lookup paths.
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py	Adds representative regression cases for base, priority, and long-context cost selection.

_{Reviews (2): Last reviewed commit: "test(pricing): cover the azure gpt-5.6 p..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing olifarhaan:litellm_azure_gpt56_pricing (506adb3) with litellm_internal_staging (c28cbb8)¹}

Footnotes

1. No successful run was found on litellm_internal_staging (e24a914) during the generation of this report, so c28cbb8 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[olifarhaan]

@greptileai tier-rate coverage added in 925c669, please take another look