Other endpoints

Besides chat completions, the API exposes three more inference endpoints. All use the same authentication, the same model field semantics, and the same error model.

Text completions

POST https://inference.inferel.ai/v1/completions

OpenAI-compatible legacy text completions (prompt in, text out). Useful for base models and code-completion workloads. Supports stream: true.

curl https://inference.inferel.ai/v1/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $INFEREL_API_KEY" \
  -d '{
    "model": "Qwen3-Coder-480B-A35B-Instruct-FP8",
    "prompt": "def fibonacci(n):",
    "max_tokens": 128
  }'

Responses API

POST https://inference.inferel.ai/v1/responses

OpenAI Responses API format, for clients built on the newer OpenAI interface:

curl https://inference.inferel.ai/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $INFEREL_API_KEY" \
  -d '{
    "model": "gpt-5.4-mini",
    "input": "Summarize the benefits of KV-cache-aware routing."
  }'

Anthropic-compatible messages

POST https://inference.inferel.ai/v1/messages

Anthropic Messages API format. Point the Anthropic SDK’s base URL at Inferel and use your Inferel key (the x-api-key header is accepted):

import anthropic

client = anthropic.Anthropic(
    base_url="https://inference.inferel.ai",
    api_key="YOUR_INFEREL_API_KEY",
)

message = client.messages.create(
    model="claude-sonnet-4.6",
    max_tokens=512,
    messages=[{"role": "user", "content": "Hello, Inferel!"}],
)
print(message.content)

Anthropic prompt caching (cache_control blocks) is supported and billed at discounted cache rates — see Prompt caching.

Image & video generation

Image and video models are served through the async job queue, not the /v1 inference endpoints — see Image & video generation.

Not supported

The API currently does not provide embeddings (/v1/embeddings) or audio endpoints. If you need these, contact support@inferel.ai.