Messages

The Anthropic-compatible Messages endpoint and its two lanes.

POST https://api.topbilling.in/v1/messages accepts the Anthropic Messages request body. Send anthropic-version: 2023-06-01.

json

{
  "model": "<model-id>",
  "max_tokens": 256,
  "system": "You are terse.",
  "messages": [{"role": "user", "content": "Three words about Chennai."}]
}

Supported

system as a string or block array (with cache_control), text / image / tool_use / tool_result blocks, tools, tool_choice, stream, stop_sequences, temperature, top_p, top_k, metadata.user_id, thinking (on reasoning models).
POST /v1/messages/count_tokens on models with token_counting.

Two lanes

When a model is served by an Anthropic-format upstream, your request bytes are relayed as-is (native lane): unknown fields and anthropic-beta values pass through. Otherwise the request is translated: unsupported semantic fields are rejected with 400 and a capability_rejected: token in the message; advisory fields such as cache_control are dropped and listed in the tb-dropped-hints response header. Each model’s page in Model Square shows which lane it uses.

Response

Standard shape: id (msg_…), content blocks, stop_reason (end_turn, max_tokens, tool_use, stop_sequence) and usage with input_tokens, output_tokens, cache_read_input_tokens, cache_creation_input_tokens.