What is Prime Intellect?
Train, deploy and improve models with RL and inference.
Who is Prime Intellect best for?
Prime Intellect is best for AI researchers who want to post-train open models with reinforcement learning.
Teams training their own agents
Hosted training runs reinforcement learning on LoRA adapters over 2,500+ community environments, with the compute and sandboxes handled for you.
Benchmarking open models
Hosted evaluations cover 100+ open-source models with no setup and a public leaderboard.
Renting GPUs by the hour
The on-demand page lists single GPUs and multi-node clusters with live hourly prices, so you can compare offers before you rent.
Researchers building RL environments
The open-source Verifiers library and the Prime CLI take an environment from first draft to evaluation and a push to the shared hub.
How much does Prime Intellect cost?
Prime Intellect charges for what you use from your account balance: hosted RL training starts at $0.02 per million input tokens, on-demand GPUs start at $0.54 per GPU per hour, and reserved clusters are priced by quote.
- Output and training tokens from $0.06 per million each
- Qwen3.6-35B-A3B: $0.25 input, $0.75 output, $1.00 training per million
- RL training of LoRA adapters on hosted environments
- Prices change in real time with supply
- Single GPUs from A6000 up to H100 and H200, or multi-node clusters
- CPU nodes from $0.06 an hour
- Offers gathered from 50+ providers
- Quotes within 24 hours
- Option to sell back idle compute
Training prices vary by model, and two models, including poolside Laguna-XS.2, are listed at $0 per million tokens. On the live GPU list, an H100 80GB started at $3.29 and an H200 at $5.48 per GPU per hour. Inference is charged per token from your balance, but Prime Intellect does not publish its per-model inference prices. Sandboxes are billed per vCPU, GiB of memory and GiB of disk for each hour they run. Full pricing on docs.primeintellect.ai. We checked these prices on .
How do readers rate Prime Intellect?
No one has rated Prime Intellect yet. Be the first.
Sign in to rate. One rating per account.
Questions about Prime Intellect
What is Prime Intellect?
Train, deploy and improve models with RL and inference.
Who is Prime Intellect for?
Prime Intellect is best for AI researchers who want to post-train open models with reinforcement learning.
Is Prime Intellect free?
No. Prime Intellect's hosted platform is pay-as-you-go: training, GPUs and inference are charged to your account balance, and no free plan is listed. Its open-source parts, such as the Verifiers library, prime-rl and the community environments on the hub, cost nothing to use.
How much does Prime Intellect cost?
Prime Intellect's paid plans are Hosted Training at From $0.02 per million input tokens and On-Demand GPUs at From $0.54 per GPU per hour.
Where can I try Prime Intellect?
On its official website, primeintellect.ai.
Epigos AI
VModel
Modal
Together AI
Modular
Anaconda