## What happened?

When using Bedrock cross-region inference models (e.g., `bedrock/us.anthropic.claude-sonnet-4-6`), the region-specific pricing entries in `model_prices_and_context_window.json` are never matched. Instead, the base model entry (`anthropic.claude-sonnet-4-6`) is always used, and region-specific pricing differences are silently ignored.

### Root Cause

The model cost lookup in `_get_model_info_helper` (`litellm/utils.py:5606`) checks candidates in this order:

1. `combined_model_name` — `bedrock/us.anthropic.claude-sonnet-4-6` → no match (no such key in cost map) 2. `model` — `bedrock/us.anthropic.claude-sonnet-4-6` → same, no match 3. `combined_stripped_model_name` — `anthropic.claude-sonnet-4-6` → **matches base model here** 4. `stripped_model_name` — `anthropic.claude-sonnet-4-6` → never reached 5. `split_model` — `us.anthropic.claude-sonnet-4-6` → **never reached**

`_strip_model_name` (`utils.py:5244`) calls `_get_base_bedrock_model` (`litellm/llms/bedrock/common_utils.py:443`), which strips both the `bedrock/` routing prefix **and** the cross-region prefix (`us.`, `eu.`, `jp.`, etc. defined in `get_bedrock_cross_region_inference_regions()` at line 411). This …