1train
Train specialist models for your tasks. Fine-tune with SFT, then reinforce with RL against rewards you define from real outcomes, on open models of any size.
Hosted Training
Train models with reinforcement learning on managed Prime-RL, our open-source trainer, using hosted compute and sandboxes.
Evaluations
Create hosted evaluations for you to benchmark the performance of your models.
Observability
Capture, view, and annotate every trace from your evals and training runs.
Sandboxes
Scale thousands of microVM sandboxes with native Docker and full RL support.
Built on open source
Prime-RL
A framework for asynchronous reinforcement learning (RL) at scale.
RLuv run rl @ examples/reverse_text/rl.tomlSFTuv run sft @ examples/reverse_text/sft.tomlInferenceuv run inference --model.name Qwen/Qwen3-0.6B
Verifiers & Environment Hub
Build RL environments with our open-source library, or start from 2,500+ community environments on the Hub.
import verifiers.v1 as vf
env = vf.SingleAgentEnv(
vf.SingleAgentEnvConfig(
taskset={"id": my_task},
agent=vf.AgentConfig(max_turns=100),
)
)
Serverless API
Dedicated deployments
Scale on the latest hardware
SLURM + K8s orchestration
InfiniBand networking
Self-healing clusters
Liquid compute
On-Demand
