Anthropic Batches support + inline requests feature

[Chesars]

Context

Issue #17265 requests direct Anthropic batch processing support.

Anthropic's inline approach

Unlike OpenAI/Bedrock (which require uploading files first), Anthropic batches accept requests inline:

# Anthropic
POST /v1/messages/batches  \
--header "x-api-key: $ANTHROPIC_API_KEY" \
   --header "anthropic-version: 2023-06-01" \
   --header "content-type: application/json" \
   --data {
  "requests": [
    {
      "custom_id": "request-1",
      "params": {
        "model": "claude-sonnet-4-20250514",
        "max_tokens": 1024,
        "messages": [
          {"role": "user", "content": "Hello, world"}
        ]
      }
    },
    {
      "custom_id": "request-2",
      "params": {
        "model": "claude-sonnet-4-20250514",
        "max_tokens": 1024,
        "messages": [
          {"role": "user", "content": "Hi again, friend"}
        ]
      }
    }
  ]
}

• No file upload
• Simpler integration
• Faster (fewer API calls)
• No file storage/cleanup to manage

Proposal

Since we're integrating Anthropic's batch API, we could offer requests inline as an alternative option alongside the existing input_file_id:

Questions

1. Add Anthropic batch support (file-based interface)? User passes input_file_id with JSONL, LiteLLM converts to inline requests for Anthropic API.
2. Also add inline requests param for other providers?
   - For OpenAI/Bedrock: LiteLLM converts inline to JSONL and uploads.
3. Input format?
   - OpenAI format (LiteLLM transforms to Anthropic)
   - Anthropic native format ( since inline is an Anthropic approach)

The inline approach doesn't follow OpenAI's file-based pattern, but it could be a valuable feature for users who prefer simpler batch workflows.