Usage-based pricing.
No subscription, no flat fees. Pay only for the runs you make — and see every cost, itemised, the moment a run finishes.
What a run costs
The benchmark
Each benchmark is priced by its complexity and the resources a run needs.
The agent
The agent that picks the right benchmark, launches the run, and analyses the results for you.
Optional
Compute
Only when you run on our GPUs. Bring your own hardware and there's nothing to pay here.
Your bill is simply the sum of these — surfaced in full after each run, so there are never any month-end surprises.
No subscription fee
Billed monthly
Itemised after every run
Optional subscriptions
Opt-in paid add-ons. Cost-saving measures carry a monthly fee but can lower per-run cost; partner tools unlock world models, simulators, and tracking. All prices are monthly.
Cost-saving best practices
- Early stopping · eval$5 / mo
- Early stopping · training$5 / mo
- Smaller agent model$3 / mo
- Compute optimization$9 / mo
- Spot / preemptible GPUs$7 / mo
- Response caching$4 / mo
- Stratified subset eval$4 / mo
Partner tools
- NVIDIA Tool Suite$49 / mo
- World Labs$39 / mo
- Odyssey$29 / mo
- Black Forest Labs$19 / mo
- Helm.ai$99 / mo
- Applied Intuition$149 / mo
- Voxel51$29 / mo
- Hugging Face Inference$25 / mo
- Weights & Biases$19 / mo
One API does everything. Our agent does it for you.
Full access is included — you only ever pay for the runs themselves.