One key gives you 53+ verified models across 19 providers — the frontier chat models, image generators, voices, transcription and decision models — billed per call from a prepaid balance you top up when you choose. Fire is what the AI Seminar cohort builds its student projects on.
Each of these is a few HTTP calls. The same key, the same balance, and the price of every step comes back in the response so you can keep an experiment inside its budget.
Record audio → POST /v1/transcribe → POST /v1/chat → POST /v1/speech.
Three calls, three call_ids, one bill. Whisper-class transcription, 5 text-to-speech engines.
Flat per-image pricing means a 100-image class demo is a known number before you start. Reference-image edits and upscaling are on the same key; a compliance-checked Layer 2 workflow handles the whole pipeline if you want it to.
Image generation →
POST /v1/chat/parallel sends one prompt to several model families at once. Rate each answer
by call_id and your ratings feed the public rankings — a real contribution, not a toy.
Every text model has a published rate per 1K tokens and every image
model a flat price per image — the same numbers on /pricing
and in GET /v1/models. Fast-tier models cost a fraction of frontier ones; pick by tier
when you are iterating and switch up for the final run.
After each call, meta.price.usd is the exact charge. The portal dashboard breaks
your spend down by model, call type and day, so a semester's budget is a chart rather than a surprise.
Every self-serve key carries every billed ability. These are the chat models live right now; the full list with prices is on /models.
By default Fire logs the metadata of a call — model, token counts, price, duration — and discards the prompt and response once the call completes. That is the same posture as the major model aggregators, and stricter than the labs' 30-day abuse-monitoring retention.
If you want to look back over your own calls, switch on private logging in the portal and
pick a retention window. Every response tells you which happened: meta.content_logged.
An account and your first API key are free. The smallest credit pack adds $5 of credit to your balance (the checkout price is a few cents more to cover the card fee, so you receive the full $5). There is no subscription and credit does not expire while your account is active.
All of them. Every self-serve key gets every billed ability — chat, image, speech, transcribe, decide and flow. There is no student tier with fewer models; the only limit is your balance.
Text models publish a per-1K-token rate and images are a flat per-image price, both listed on /pricing and in GET /v1/models. After the call, meta.price.usd in the response is the exact amount charged, and the portal dashboard shows spend per model and per day.
Not unless you opt in. Fire keeps metadata — which model, how many tokens, the price — and discards prompt and response content after the call by default. You can switch on private logging in the portal if you want to review your own calls, and choose how long they are kept.
No. Any HTTP client works, and the examples are plain curl. If you prefer, there are packages for Python (fire-sdk on PyPI), JavaScript (@moduspromethean/fire-sdk on npm) and PHP.
Create an account, mint a key, read the catalog. Top up when you have something to build.