Open models, kept running.

When a big host retires a model your product depends on, Anix keeps it online: the same weights, behind the same model ID, on an OpenAI-compatible API.

Request an API key Read the quickstart

Models and pricing

Prices are in US dollars per million tokens. The cached rate applies when a request repeats a prompt prefix the server has recently processed, which is most turns of an agent session.

Model	Status	Context	Precision	Input	Cached input	Output
Laguna XS 2.1poolside/laguna-xs-2.1	Running	262K	FP8	$0.10	$0.05	$0.20
Laguna S 2.1poolside/laguna-s-2.1	Coming soon	–	–	–	–	–

Depending on a model another host just dropped? Tell us which one.

Quickstart

Any OpenAI-compatible client works. Point it at our base URL and use a model ID from the table above.

With curl

curl https://api.useanix.com/v1/chat/completions \
  -H "Authorization: Bearer $ANIX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "poolside/laguna-xs-2.1",
    "stream": true,
    "messages": [
      {"role": "user", "content": "Write a binary search in Rust"}
    ]
  }'

With the OpenAI Python SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.useanix.com/v1",
    api_key="YOUR_ANIX_API_KEY",
)

reply = client.chat.completions.create(
    model="poolside/laguna-xs-2.1",
    messages=[{"role": "user",
               "content": "Write a binary search in Rust"}],
)
print(reply.choices[0].message.content)

Base URL: https://api.useanix.com/v1

What every request gets

The model you built onThe maker's published weights, at the precision shown above, under the same model ID.
OpenAI-compatible chat completionsStreaming, tool calls and reasoning output, with usage reported on every response.
Zero data retentionPrompts and outputs are never logged, stored or used for training. See how requests are handled.
Cheaper long sessionsRepeated prompt prefixes are cached and billed at the cached rate.
Hosted in the United StatesAll inference runs on GPUs in the US.
Notice before anything changesIf we ever retire a model you've used in the last 30 days, you get at least 30 days' notice by email.

Why Anix exists

Big hosts retire models to free their GPUs for newer ones. If your product, prompts or evals were tuned to one of those models, its retirement forces a migration on someone else's schedule. Anix keeps those models running, so you move when you're ready.

Dedicated endpoints

Need reserved capacity, a model version pinned for as long as you need it, or your own Laguna fine-tune served behind an API? We run dedicated endpoints on GPUs reserved for you, for a flat monthly price.

Ask about a dedicated endpoint