[Test Fix] fix gov pricing tests

[clfhhc]

Relevant issues

test_govcloud_completion_cost_calculation
test_govcloud_completion_with_cost_tracking
test_govcloud_cost_per_token_with_region
test_govcloud_model_cost_properties
test_govcloud_model_pricing_verification
test_govcloud_models_in_model_cost

mypy_linting

Pre-Submission checklist

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

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Delays in PR merge?

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

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

• Fix cost-related pricing for gov cloud
  https://app.circleci.com/pipelines/github/BerriAI/litellm?branch=litellm_fix_gov_pricing_tests

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Apr 2, 2026 10:53pm

[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.}

[codspeed-hq]

Merging this PR will not alter performance

✅ 16 untouched benchmarks

---

_{Comparing litellm_fix_gov_pricing_tests (797c747) with main (06df8ed)}

[greptile-apps]

Greptile Summary

This PR fixes failing govcloud pricing tests by: (1) adding None-safe guards (or 0.0) throughout cost calculators to handle model entries whose pricing fields exist in the JSON but have a null value; (2) adding new bedrock/us-gov-east-1/anthropic.claude-haiku-4-5-20251001-v1:0 and bedrock/us-gov-west-1/anthropic.claude-haiku-4-5-20251001-v1:0 pricing entries; (3) correcting govcloud Claude Sonnet 4.5 pricing (3.3e-06→3.6e-06 input, 1.65e-05→1.8e-05 output); and (4) aligning govcloud tests to use anthropic.* base model IDs rather than us.* inference-profile IDs, which matches actual Bedrock GovCloud API behavior.

Key changes:

• litellm/cost_calculator.py, router.py, llms/azure_ai/cost_calculator.py, llms/vertex_ai/cost_calculator.py, litellm_core_utils/llm_request_utils.py: defensive or 0.0 / truthiness guards preventing TypeError when a model's pricing field is None
• model_prices_and_context_window.json + backup: two new govcloud Haiku 4.5 entries; corrected Sonnet 4.5 govcloud pricing
• tests/llm_translation/test_bedrock_govcloud.py: model IDs corrected to anthropic.* for GovCloud, region_name removed from base-model cost calls
• tests/logging_callback_tests/langfuse_expected_request_body/completion_with_bedrock_call.json: totalCost changed from 6.599999999999999e-05 to 6e-05 — the old value was consistent with us.* inference-profile pricing while the new value matches the cheaper base anthropic.* pricing, warranting verification

Confidence Score: 4/5

Safe to merge pending clarification of the totalCost change in the Langfuse fixture, which may indicate cost tracking underreports for us.* inference-profile Bedrock models.

The core govcloud fixes are well-reasoned and the None-guard changes are correct. One P1 finding remains: the Langfuse fixture update changes the expected total cost to a value consistent with cheaper base model pricing rather than the us.* inference-profile pricing declared in the JSON, which contradicts an assertion in the govcloud test suite itself.

tests/logging_callback_tests/langfuse_expected_request_body/completion_with_bedrock_call.json — the totalCost value should be verified against the actual bedrock cost-lookup path for us.* inference profiles.

Important Files Changed

Filename	Overview
litellm/cost_calculator.py	Adds or 0.0 guard for None pricing values to prevent TypeError on None > 0 comparisons.
litellm/litellm_core_utils/llm_request_utils.py	Same None-safe guard in the cheapest-model picker; minor formatting inconsistency introduced.
litellm/llms/azure_ai/cost_calculator.py	Adds truthiness check before comparing router_flat_cost_per_token > 0, preventing TypeError when value is None.
litellm/llms/vertex_ai/cost_calculator.py	Adds or 0.0 fallback when multiplying by model_info["input/output_cost_per_token"] to handle None values.
litellm/router.py	Wraps cost comparisons in (... or 0.0) to guard against None; outer is not None check makes the guard on model_info side redundant but harmless.
model_prices_and_context_window.json	Adds two new govcloud Haiku 4.5 entries; corrects Sonnet 4.5 govcloud pricing from 3.3e-06/1.65e-05 to 3.6e-06/1.8e-05.
tests/llm_translation/test_bedrock_govcloud.py	Fixes test model IDs to use anthropic.* for GovCloud (not us.*); all tests remain properly mocked with no real network calls.
tests/logging_callback_tests/langfuse_expected_request_body/completion_with_bedrock_call.json	Expected totalCost changed from 6.6e-05 (consistent with us.* profile pricing) to 6e-05 (consistent with base anthropic.* pricing) — potential pricing lookup regression.

_{Reviews (3): Last reviewed commit: "fix test get model info" | Re-trigger Greptile}

[greptile-apps]

Silent $0 cost when pricing is None

Both input_cost_per_token and output_cost_per_token are accessed via model_info["..."] (dict key access that raises KeyError on missing keys) but the or 0.0 fallback silently produces a $0 cost when those keys exist with a None value. If a Vertex AI model somehow has input_cost_per_token: null in the pricing JSON, the cost calculation will silently undercharge users instead of surfacing the missing data.

The govcloud models added in this PR all have explicit non-None pricing, so this won't trigger here. Consider raising a more explicit error or logging a warning when the value is None so it's not silently ignored:

_input_cost = model_info["input_cost_per_token"]
if _input_cost is None:
    verbose_logger.warning("input_cost_per_token is None for model_info, defaulting to 0.0")
    _input_cost = 0.0
prompt_cost = prompt_tokens * _input_cost

[clfhhc]

makes sens but I'm not sure what our mental model is

[ishaan-berri]

lgtm

[greptile-apps]

Expected cost inconsistent with us.* inference-profile pricing

The totalCost changed from 6.599999999999999e-05 to 6e-05. The model logged is bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0 with 10 input + 10 output tokens.

Pricing cross-check from model_prices_and_context_window.json:

• us.anthropic.claude-haiku-4-5-20251001-v1:0 → input: 1.1e-06, output: 5.5e-06 → 10 × 1.1e-06 + 10 × 5.5e-06 = 6.6e-05 (old value — correct for the us.* profile)
• anthropic.claude-haiku-4-5-20251001-v1:0 (base Bedrock) → input: 1e-06, output: 5e-06 → 10 × 1e-06 + 10 × 5e-06 = 6e-05 (new value — matches base-model pricing, not the us.* profile)

The new fixture documents that bedrock/us.anthropic.* is now being priced via the cheaper base anthropic.* key, underreporting cost by ~9%. This is also internally inconsistent with test_govcloud_completion_cost_calculation, which still asserts us.* profile pricing for the same model string:

expected_base_cost = 10 * 1.1e-06 + 5 * 5.5e-06
assert abs(base_cost - expected_base_cost) < 1e-10

Please confirm which pricing key the bedrock cost calculator resolves for us.* inference-profile model strings and verify this fixture was not regenerated from a regressed code path.

Rule Used: What: Flag any modifications to existing tests and... (source)