Model API
Open models that keep getting faster
Pay per token with zero data retention, and get every speedup we ship automatically.
Self-improving
An API that gets faster the more it's used
More traffic gives our agents more to learn from. They turn it into faster serving recipes, prove them and ship them to everyone, then start again.
Your traffic makes it faster
Our agents study the shape of every request, its timing and length but never its text, to find where each model can run faster.
Only proven changes reach you
Candidate serving recipes are replayed against that shape, then run on a shadow copy of live traffic before anything ships.
Same model, just faster
A recipe ships only when it's faster at the same quality, and every user gets it at once, with nothing to change.
Privacy
Zero data retention on every request
We never store, log or train on your prompts or outputs, and there's nothing to switch on. Billing keeps only what it needs.
- Never stored, logged or trained on
- Your prompts and outputs
- Kept for billing
- Token counts, model, timestamp and cost
- Regions on request
- The US, the EU or Asia
Speed
The fastest on Vercel AI Gateway
Vercel measures every provider's speed in public, model by model. Right now, we're #1 on every model shown.
Kimi K3
#1 of 15
Median output speed, last hour
- 1Boundless280 tok/s
- 2Nebius278 tok/s
- 3Runware193 tok/s
- 4Parasail191 tok/s
- 5Modal124 tok/s
Cost
Pay less per token than at the major providers
Pay per token from prepaid credits, with no fee or commitment.
No surprise bills
Requests stop when your credits run out, so a runaway job can never spend past them.
Spend limits per team
Give each app, team or environment its own key and budget, so their spend stays separate.
Repeated prompts cost less
Repeated input, such as a system prompt, is billed at a lower cached rate.
The base URL
Point your SDK at Boundless. For Anthropic SDKs, drop the /v1.
The API key
Sign up in the console and create a key. No sales call.
The model name
Pick any model in the library.
curl https://api.inference.boundless.network/v1/chat/completions \
-H "Authorization: Bearer $BOUNDLESS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-v4.1-flash-1017",
"messages": [{
"content": "Say hello",
"role": "user"
}]
}'FAQ
Questions before you build
Can we switch without rewriting our code?
Yes. The Boundless API works with the OpenAI and Anthropic SDKs, so switching means changing the base URL, the API key, and the model name.
Where does our data go?
Every request runs with zero data retention, so prompts and outputs are never stored, logged or used for training. If your data has to stay in one region, you can request inference in the US, the EU or Asia.
Do we get speed improvements automatically?
Yes. Every improvement goes live on the endpoint you already call, with nothing to migrate and nothing extra to pay. It changes how fast and reliably a model responds, never the model or its quality.
How does billing work?
You pay per token from prepaid credits. Requests stop when credits run out, so there are no surprise bills, and you can set a budget for each API key.
What's the difference between the API and tailored inference?
The API serves open models we've tuned at launch, billed per token. With tailored inference, our engineers tailor the model and serving to your traffic and keep optimizing it 24/7.
Can AI agents pay without an API key?
Yes. Over MPP, each request is charged for the tokens the model actually used, at the same per-token prices as our API, and failed requests are refunded. How MPP works
How do we get started?
Sign up and use your free credits to try the API today, or talk to an engineer about a free evaluation on your traffic.




