8 views
-/https://github.com/berriai/litellm/issues/30768
GitHub · issue

#30768 [Bug]: Incorrect cost tracking for Bedrock geo cross-region inference profiles (us./eu./ap.)

  • State: open
  • Author: @Prathamesh010
  • Labels: bug, proxy, llm translation

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

When a model is configured with a Bedrock cross-region inference profile ID (e.g. us.anthropic.claude-sonnet-4-6), LiteLLM's get_model_info resolves pricing using the base-model key (anthropic.claude-sonnet-4-6) instead of the cross-region key (us.anthropic.claude-sonnet-4-6). Cross-region inference profiles carry a ~10 % surcharge, so the reported/logged cost is understated.

<img width="1413" height="52" alt="Image" src="https://github.com/user-attachments/assets/4ca475ca-ec3e-4130-88ec-b7d20bfe6412" />

**Expected behavior**

get_model_info("bedrock/us.anthropic.claude-sonnet-4-6") should return the cross-region pricing:

input_cost_per_token: 1.1e-06 ($1.10 / 1M tokens) output_cost_per_token: 5.5e-06 ($5.50 / 1M tokens)

**Actual behavior**

It returns the base-model pricing:

input_cost_per_token: 1e-06 ($1.00 / 1M tokens) output_cost_per_token: 5e-06 ($5.00 / 1M tokens)

### Steps to Reproduce

use proxy config: Proxy config ``` model_list: - model_name: Claude Sonnet litellm_params: model: bedrock/us.anthropic.claude-sonnet-…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (5 events)
#0 of 0 · 31d18h49m6s ago — entered · #import:https:::github.com:berriai:litellm post #2255
Preserving provider-specific metadata requires tracing usage through translation, aggregation, streaming, response schemas, and compatibility tests, creating broader integration and backward-compatibility risk. The pricing issue is more localized to model identification and cost-data resolution with focused regression coverage.
The right issue requires tracing and modifying runtime pricing-resolution logic, handling provider-specific identifiers and regression coverage. The left issue is a contained metadata update across two synchronized files.
The left issue requires tracing model reconstruction across database persistence, provider-specific configuration selection, and the streaming request path, with compatibility and regression risks across multiple registration mechanisms. The right issue is comparatively localized to pricing-key resolution and associated cost-calculation tests.
The right-hand task has broader provider-specific lookup and pricing implications, with higher regression risk across model aliases and cost calculation paths. The left-hand task is comparatively localized to aligning UI authorization checks with an already-supported backend permission.
#0 of 0 · 31d17h51m10s ago — current · #import:https:::github.com:berriai:litellm post #3230
The right issue is harder because it involves provider-specific model and pricing resolution, preserving accurate cost calculations across Bedrock inference-profile variants, and validating behavior across shared pricing paths. The left issue is comparatively localized to empty-value handling between the dashboard and budget-update validation.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search