Turion Inference runs open ML models — speech-to-text, image generation, and your own Cog-packaged models — on dedicated GPUs behind a simple key-authed API, with a Python SDK and CLI. Now in private alpha.
$ export TURION_API_TOKEN=tk_...
$ turion run whisper -i audio=@meeting.wav
[06a3b2] status=starting
[06a3b2] status=processing
[06a3b2] status=succeeded
{
"text": "And so my fellow Americans, ask not what your
country can do for you...",
"language": "en"
}import turion
client = turion.Client() # reads TURION_API_TOKEN
out = client.run("whisper", input={"audio": "meeting.wav"})
print(out["text"])
art = client.run("illustrious-xl", input={
"prompt": "(masterpiece)1.2, watercolor lighthouse at dawn",
"width": 832, "height": 1216,
})
print(art["images"][0]) # hosted result URLTell us what you want to run. We onboard in small batches so every alpha user gets real capacity, not a queue.
You receive a key and a base URL. Auth is a bearer token — no OAuth dance, no console required.
Call the REST API directly, or use the Python SDK and CLI. File inputs upload automatically; results come back as JSON and hosted URLs.
OpenAI's Whisper, rebuilt on current PyTorch for our GPUs. Send audio, get a transcript with language detection — validated end-to-end on the same hardware that serves the alpha.
SDXL-class illustration checkpoint with the full parameter set — prompt weighting, PAG, CLIP skip, scheduler choice. Returns hosted image URLs.
The platform runs Cog-packaged models — the same container format Replicate uses. During the alpha we onboard additional models per request; tell us what you need on the waitlist form.
We onboard in small batches and will email you when yours opens. Meanwhile, the blog covers what we're building.