Compatibility matrix · verified routes, not guessed capabilities

Choose the route before you change the model.

AIWave keeps one key, one USD balance, and one domain. It does not pretend that OpenAI and Anthropic use the same wire protocol.

Compare model families Configure a client

Model-family request contract

A shared account removes commercial re-integration. The request and response shape still follows the model family.

Family	Recommended route and client	Output limit	Sampling fields	Streaming response	Last verified
Chinese models	`/v1/chat/completions` OpenAI-compatible client	`max_tokens` in current quickstart examples; verify model-specific exceptions	`temperature` and `top_p` are model-specific	OpenAI-compatible SSE chunks on documented routes	2026-09-24 tool-call matrix; streaming coverage remains partial
GPT	`/v1/chat/completions` Existing OpenAI client	Use `max_completion_tokens` when the model rejects `max_tokens`; current GPT route applies its verified parameter rule	Do not force a non-default value unless the selected model accepts it	OpenAI-compatible shape; streaming not separately certified for every GPT model	2026-09-28
Claude	`/v1/messages` Anthropic SDK or native HTTP	`max_tokens`	The verified route removes deprecated `temperature` and `top_p` fields before upstream delivery	Anthropic event stream; Chatbox path verified, every native event type not yet certified	2026-10-01

Honest boundary.

A Claude request sent to /v1/chat/completions can be accepted and billed, but it returns an Anthropic envelope. No OpenAI choices field is present, so an OpenAI SDK parser fails. Use /v1/messages instead.

Claude: keep the account, use the native client.

The key, balance, and domain stay the same. Only the protocol-specific client changes.

Anthropic Python SDK

# Set ANTHROPIC_API_KEY to your AIWave key before running this example.
from anthropic import Anthropic
client = Anthropic(base_url="https://aiwave.live")
message = client.messages.create(model="claude-haiku-4-5-20251001", max_tokens=64, messages=[{"role": "user", "content": "Reply with exactly: OK"}])
print(message.content[0].text)

Native HTTP

# Export AIWAVE_API_KEY in your shell before running this request.
curl https://aiwave.live/v1/messages \
  -H "x-api-key: $AIWAVE_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-haiku-4-5-20251001","max_tokens":64,"messages":[{"role":"user","content":"Reply with exactly: OK"}]}'

Examples use placeholders. Controlled route verification is recorded separately; never paste a real key into shared code or screenshots.

Known failure modes

Match the symptom before changing credentials or routes.

Symptom	Cause	Fix
OpenAI client reports a missing `choices` field for Claude	The Claude route returned its native Anthropic envelope	Use the Anthropic SDK or POST `/v1/messages`
`max_tokens` is unsupported on a GPT model	The model expects `max_completion_tokens`	Send `max_completion_tokens`; retain the upstream error body when reporting a recurrence
Claude reports that `temperature` or `top_p` is deprecated	The model requires those fields to be omitted	The verified AIWave Claude route removes both fields; if the error returns, keep the request ID and contact support
A stream parser receives unfamiliar event types	An Anthropic event stream was parsed as OpenAI SSE chunks	Use the matching Anthropic stream iterator for Claude

Run one bounded check before moving a workload.

Start with the quickstart, copy the exact fields from client setup, then compare the returned shape with this matrix.