90+ models from 12 AI labs, plus the live web, transcription and images, all behind one key. Flux sends every job to the best model for it, and shows you what each one cost.
One key, any agent.
If it can talk to OpenAI or Anthropic, it can talk to Flux. Including your own custom endpoints.
MurageMurage asks
Read today’s Google CalendarShipping$0.010
Scan Gmail for anything urgentShipping$0.010
Examples. Names, prices and costs are illustrative. How Murage did “Morning brief”
Any model can tell your agent what to do. Flux is what lets it actually do it, for less, on whichever model is best this week.
A brain alone: a suggestion
With Flux: the job, finished
A brain on its own gives you an answer. With the live web, images and more on the same key, your agent does the work as well as the talking.
About 85% less in this example. Illustrative figures, not a measurement.
Flux uses a frontier model where it matters and a fast one where it doesn't, so every step costs what it should.
Every call comes back with the model that ran it and what it cost, all on one bill.
Flux picks the right one for each job, or you pin your favourite.
Search, then read the pages that matter.
Search: rolling out · Page reading: live
61 voices to speak with, and transcription to listen.
Speak: rolling out · Transcribe: live
Gmail, Slack, GitHub, Notion and hundreds more, with permission.
Generate a picture, then keep refining it.
What each user tells your agent stays with them, whichever model answers.
These aren't six products.
It's one key.
Every model lab and every tool wants its own account, key, bill and integration. Flux replaces the pile with one key, and routes each job to whichever option does it best.
5 vendors to set up, pay and keep working.
One key in the tool you already use.
FLUX_API_KEY=sk-YOUR-FLUX-KEY
All of it works with any major model, from the agent you already use.
90+ models from 12 AI labs. Flux reads each job and picks the best fit, or you pin the one you want.
Search the web and read clean page text, so answers use what's true today.
61 voices to speak with, plus transcription for anything your agent hears.
Let your agent act in Gmail, Slack, GitHub, Notion and hundreds more, with each user's permission.
Generate an image, then keep editing it until it's right. About 24 image models behind one name.
What each user tells your agent stays with them across sessions and models. Every fact shows where it came from, and deleting it really deletes it.
For answers that matter, Flux checks the answer with other models before you act. It stops at a fixed cost cap.
Checked in September 2026, including where the others win.
| Capability | Flux | OpenRouter | Vercel AI Gateway | Going direct |
|---|---|---|---|---|
| Every major model on one key | Yes | Hundreds | Yes | One per vendor |
| Picks the right model for each job | Yes | Auto Router | You choose | No |
| Live web search and page reading | Page reading now | Yes | Yes | Some |
| Voices, speaking and listening | Transcription now | Yes | Yes | Per vendor |
| Actions in apps | Coming to your key | No | No | No |
| Memory across every model | Shipping | No | No | Per vendor |
One key, any agent. Point the tool you already use at Flux and keep everything else.
