litellm - 💡(How to fix) Fix Llama 4 Maverick/Scout on Bedrock: tool calling fails with "provider rejected the request schema or tool payload"

Environment

• LiteLLM version: 1.85.0
• Model: us.meta.llama4-maverick-17b-instruct-v1:0 (cross-region inference profile)
• Provider: AWS Bedrock Converse API
• LiteLLM config: bedrock/converse/us.meta.llama4-maverick-17b-instruct-v1:0

Steps to reproduce

1. Configure LiteLLM proxy with llama4-maverick via bedrock/converse/us.meta.*
2. Send a request that includes any tool/function definition
3. Request fails immediately before the model generates any output

Error

LiteLLM returns HTTP 500 with: litellm.APIConnectionError: BedrockException - {"Message":"..."}

The calling application (OpenClaw) reports: FailoverError: LLM request failed: provider rejected the request schema or tool payload.

Root cause hypothesis

Bedrock enforces strict JSON Schema draft 2020-12 validation on tool definitions for Llama 4 cross-region inference profiles. The tool schema format LiteLLM sends for openai-completions-compatible tool calls may not pass Bedrock's validation.

Interestingly, Nova models (amazon.nova-micro-v1:0) accept the same tool definitions without issue — this appears specific to Llama 4 cross-region inference profiles.

Expected behavior

Tool calls should work with Llama 4 Maverick/Scout the same as with Nova models.

Workaround

Currently using OpenRouter as fallback for tool-enabled agent runs. Would prefer to use Bedrock Llama 4 to avoid rate limiting.