llmfa.st

Fast, cheap inference
for open models.

We run Qwen, DeepSeek, GLM and Kimi on our own GPUs and hand you back tokens the moment they exist. One OpenAI-compatible API. No queue.

Get started See the models

The whole idea, on one napkin

You already know how to call an OpenAI-compatible endpoint. Point it at us instead and the same code runs against open weights, on hardware we own.

Why it's quick

Your system prompt is computed once

Prefix caching keeps the shared start of your conversations resident on the GPU. Repeat calls skip straight past it — and we bill those cached tokens at a fraction of the normal rate.

We never make you wait in line

When a GPU is full we return a 429 immediately instead of holding your request. Queueing looks like speed on our dashboard and feels like slowness in your app. We'd rather refuse than pretend.

Nothing buffers on the way out

Every token is flushed to your socket the moment it exists — no batching, no compression, no proxy sitting on the stream. First token out is first token seen.

Models

Open weights, run on our own metal. Every model keeps the licence its authors published it under.

Qwen3 Alibaba textvisiontools

DeepSeek DeepSeek AI textreasoningtools

GLM Z.ai textreasoningtools

Kimi Moonshot AI textlong contexttools

The exact models we serve, their context windows, capabilities and current pricing are always at api.llmfa.st/v1/models. That endpoint is the source of truth; nothing on this page is. You can also compare us against every other provider on OpenRouter.

Get started

Change two lines. If your code already talks to an OpenAI-compatible API, it already talks to us.

# Python — the OpenAI SDK, pointed at us
from openai import OpenAI

client = OpenAI(
    base_url="https://api.llmfa.st/v1",
    api_key="sk-llmfast-...",
)

stream = client.chat.completions.create(
    model="qwen/qwen3-32b",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Or reach us through OpenRouter, where we sit alongside every other provider and you can compare us on price and speed before sending a single request.

Request an API key

Questions? Just email us — a person reads it.

Fast, cheap inference for open models.