The free inference network

An OpenAI-compatible endpoint with no API key. What it is, what it costs, and why the first call is slow.

Animica runs an OpenAI-compatible inference endpoint at https://animica.dev/v1. There is no API key:

curl https://animica.dev/v1/chat/completions \
  -H 'content-type: application/json' \
  -d '{
    "model": "animica-chat-small",
    "messages": [{"role": "user", "content": "Explain bech32m in two sentences."}]
  }'

Because it is OpenAI-shaped, any OpenAI client library works — point the base URL at https://animica.dev/v1 and pass any string as the key.

Check which models are actually serving. GET /v1/models returns a serving flag per model. Measured on 2026-08-20:

model	serving
`animica-chat`	yes
`animica-chat-small`	yes
`animica-chat-flagship`	no

A model with serving: false will queue behind a cold start or return 503. Requesting a model id that is not on the list at all is worse: you may get a reply that ignored your prompt entirely. Read the list; do not assume.

Expect ~20 seconds, not ~2. Measured latency for a short completion on 2026-08-20 was 21 seconds, and a cold start is longer. This is community GPU capacity donated by miners, not a hyperscaler.

Design around it: never block a page render on a completion. Stream, show progress, or do the work in the background and cache the result. The rate limit is 30 requests per minute per IP.

Stuck? Ask

Answered by Animica's own free inference network. It is donated GPU capacity, so give it 20-30 seconds.

The free inference network

Claim your 10 ANM

Stuck? Ask