LiteLLM Proxy: One model_list, Both OpenAI and Anthropic Endpoints

LiteLLM Proxy exposes the OpenAI-compatible /v1/chat/completions route, but it also natively implements the Anthropic /v1/messages route. That route can forward to any backend LiteLLM supports — OpenAI, Azure, Bedrock, Vertex — not just Anthropic models. In other words, a client that only speaks the Anthropic wire format can drive a non-Claude model.

Why this matters

Claude Code always sends requests in Anthropic’s format. Pointing it at LiteLLM Proxy lets it call, say, GPT-4o, while Claude Code itself never knows the backend changed.

# config.yaml
model_list:
  - model_name: claude-opus-4-7      # name exposed to clients (select this in Claude Code)
    litellm_params:
      model: openai/gpt-4o           # actual backend model
      api_key: os.environ/OPENAI_API_KEY

litellm_settings:
  master_key: os.environ/LITELLM_MASTER_KEY

Start the proxy:

litellm --config config.yaml   # listens on http://0.0.0.0:4000

Point the client at it:

export ANTHROPIC_BASE_URL="http://0.0.0.0:4000"
export ANTHROPIC_AUTH_TOKEN="$LITELLM_MASTER_KEY"
claude --model claude-opus-4-7   # requests are forwarded to openai/gpt-4o

Or hit /v1/messages directly with curl:

curl -X POST http://0.0.0.0:4000/v1/messages \
  -H "Authorization: Bearer $LITELLM_MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-opus-4-7","max_tokens":100,"messages":[{"role":"user","content":"hi"}]}'

The request body is Anthropic-shaped. LiteLLM translates messages, tool calls, and streaming into whatever the backend provider (OpenAI in this case) expects, then translates the response back to Anthropic’s format on the way out.

Both endpoints, same model_list, no extra config

/v1/chat/completions and /v1/messages are two protocol “shells” over the same backend model list — nothing extra needs to be configured to use both at once.

Using the claude-opus-4-7 → openai/gpt-4o mapping above, both calls work against the same deployment:

# OpenAI format
curl -X POST http://0.0.0.0:4000/v1/chat/completions \
  -H "Authorization: Bearer $LITELLM_MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-opus-4-7","messages":[{"role":"user","content":"hi"}]}'

# Anthropic format
curl -X POST http://0.0.0.0:4000/v1/messages \
  -H "Authorization: Bearer $LITELLM_MASTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-opus-4-7","max_tokens":100,"messages":[{"role":"user","content":"hi"}]}'

Both requests target the same model_name and land on the same backend model. Only the request/response wire format differs:

This is LiteLLM’s core value as a gateway: one model_list, multiple client protocols, interchangeable backends.

Reference

• Anthropic API - LiteLLM docs