About Inference APIs
One OpenAI-compatible endpoint for popular open models — chat, speech and transcription — billed per request, with no tiers to unlock and no daily caps to outgrow.
What we do
We serve 21 production models today across chat, text-to-speech and speech-to-text, all behind the same base URL and API key. Requests use the OpenAI wire format, so the official OpenAI SDKs, LangChain, LiteLLM and most agent frameworks work by changing the base URL. Widely used ids from other providers (for example llama-3.3-70b-versatile or whisper-1) are accepted as aliases so migrations are configuration changes, not rewrites.
Why it exists
Free tiers cap out, upgrade paths close, and providers retire models. In 2026 we watched thousands of small projects break when Groq retired Llama 3.3 70B and paused Developer-tier upgrades. Inference APIs exists for the developer who has a working app and needs a straightforward, metered endpoint that keeps working — the same model id, a different base URL, and a bill that scales with use.
How it is built
- Upstream capacity. Chat, speech and transcription models are served from Together AI's serverless fleet, with our gateway handling authentication, metering, aliasing and error normalisation. Vision parsing (OmniParser) runs on our own GPU hardware.
- Pricing. Every model is priced at a fixed margin over our upstream cost. Rates are published per model on the pricing page and change only when upstream costs change.
- Data. Requests are proxied and not stored beyond what is needed to meter them (model, token/character/minute counts, cost). Prompt and audio content is not retained by us after the response is returned, and zero data retention is switched on with the GPU providers that run the models, so they do not keep it either. What our upstream providers keep, and at what precision each model runs, is set out per model on the trust page.
Who runs it
Inference APIs is built and operated by Taylor Hawkes, an independent software engineer. It is a small operation on purpose: one person answers support email, writes the docs and the reference, and is on the hook when something breaks. If that matters to you, welcome; if you need an enterprise SLA and a sales team, we will happily point you elsewhere.
Contact
Email support@inferenceapis.com, or see the contact page. Current availability is on the status page; what changed and when is in the changelog.