The Open Superintelligence Stack Own Your Intelligence Own Your Intelligence Own Your Intelligence Train, deploy, and continuously improve your own models on an integrated compute, training, inference, and sandbox stack. START TRAINING BOOK A CALL $ pip install prime pip install prime Backed by Founders Fund / Radical / NVIDIA / Intel / Andrej Karpathy / John Schulman / Dylan Patel / Clem Delangue Case study Read more Case study Read more Lab. Post-train your own self improving agents Start Training Book a Demo FIG.1 01 RL Environments Turn any task into an RL environment. Init, develop, eval, and push with the Prime CLI. 1.1 Built on the open-source Verifiers library 1.2 One CLI loop: init, develop, eval, push 1.3 2,500+ community environments on the Hub Create Environments FIG.2 02 Evaluations Hosted evaluations for you to benchmark the performance of your models. 2.1 100+ open-source models 2.2 No infra, no setup. 2.3 Public leaderboard Run your first eval FIG.3 Reward 0.01 1 0.8 0.6 0.4 0.2 0 max_steps 10,000 rollouts_per_example 19 seq_len 4 batch_size 65536 max_tokens 256 learning_rate 0.00005 03 Hosted Training Train large-scale models optimized for agentic workflows. 3.1 Train on 2,500+ RL environments 3.2 Managed training workflows with full visibility and control 3.3 Hands on support from our applied research team Start Training FIG.4 04 Inference Dedicated or serverless inference for your custom models, with native LoRA support. 4.1 1-click deployment for any fine-tuned model 4.2 LoRA adapters served alongside base models 4.3 Zero config. No setup. Book a Call “We worked with Prime Intellect to train Fast Ask on Lab — a small RL-trained subagent that helps the Ramp Sheets agent find answers inside spreadsheets. The result beat the frontier models on accuracy while running at faster speeds and a fraction of the cost. Rather than wait on a better frontier model, we trained our own for the workflow that mattered to us” Karim Atiyeh Ramp Co-CEO “Evals are the foundation for building better agents. Prime Intellect helps turn them into real improvement loops.” Robin Salimans Principal AI Engineer Environment Hub Access and contribute to 2,500+ open-source RL environments and a community of researchers and developers. Explore Environments Explore My Stars My Environments Featured 9 Show All primeintellect 2 opencode-science Solve science problems using OpenCode agent via... science opencode +1 Updated 8 days ago v0.3.8 primeintellect 6 deepdive DeepDive QA RL environment with a Serper-powered search tool rl qa +1 Updated 11 days ago v0.2.5 stochi0 3 rubric-discovery Meta-environment for learning rubric functions from labeled... rlm training +4 Updated 2 months ago v0.2.0 INTELLECT-3 3 primeintellect 8 mini-swe-agent-plus Mini SWE Agent Plus environment for solving SWE issues inside Pri... swe sandbox +1 Updated 3 days ago v0.2.23 primeintellect 6 deepdive DeepDive QA RL environment with a Serper-powered search tool rl qa +1 Updated 11 days ago v0.2.5 primeintellect 3 science-env A collection of challenging single-turn science problems science single-turn Updated 11 days ago v0.1.3 Evals 13 Show All hud 18 hud-text-2048 Text-based 2048 game for training agents to reach target tiles through strategic moves game text +2 Updated 7 months ago v0.1.0 hud 18 hud-text-2048 Text-based 2048 game for training agents to reach target tiles through strategic moves game text +2 Updated 7 months ago v0.1.0 will 29 will/tau2-bench Verifiers implementation of tau2-bench tool-agent-user tool-use +2 Updated 2 months ago v0.1.0 Verifiers 1 2 3 4 5 6 7 8 9 import verifiers as vf vf_env = vf.ToolEnv( dataset=dataset, parser=parser, rubric=rubric, tools=tool_list, max_turns=10, ) A library of modular components for creating RL environments and training LLM agents. Prime-RL uv run rl \ --trainer @ examples/reverse_text/ rl/train.toml \ --orchestrator @ examples/ reverse_text/rl/orch.toml \ --inference @ examples/ reverse_text/rl/infer.toml A framework for asynchronous reinforcement learning (RL) at scale. Sandboxes deepswe-sandbox-1 python:3.11-slim deepcoder-sandbox-1 python:3.11-slim i3-math-sandbox-1 python:3.11-slim For secure code execution optimized for large-scale reinforcement learning. Inference. Serve open-source and custom models through the same stack that trains them. Book A Call Learn More Dedicated inference, pay-per-token LoRA serving, and serverless APIs-all in one loop that turns traces into better models and cheaper intelligence your business owns. Dedicated deploys Production serving capacity optimized around customer use cases, private routing, latency, reliability, and custom model requirements. LoRA inference Pay-per-token serving for adapters trained with Lab, so teams can deploy customized behavior without copying the base model. Serverless APIs OpenAI-compatible access to base models with efficient routing, selected first-party hosting, and a path into dedicated capacity. Turn production traces into the next training run. Capture traces, cluster failures, convert high-value misses into environments and evals, then train adapters that make the production model cheaper, more reliable, and more specific to your business. Book a Call Compute. Find reliable compute operated globally from a single GPU to largest clusters. FIND COMPUTE BOOK A DEMO On demand Instant access to 1-256 GPUs. Use your GPUs across clouds in a single platform. 1.1 SLURM, K8s Orchestration Orchestrate dynamic workloads with enterprise-grade scheduling and container automation. 1.2 Infiniband Networking Scale distributed training with high-bandwidth interconnects across nodes. 1.3 Grafana Monitoring Dashboards Visualize metrics in real time with customizable dashboards for full system observability. GET COMPUTE FIG.5 Single-Node Multi-Node H200 Available x2 · x1 $1.99/HR 80 GB VRAM · 184 GB RAM · 32 vCP H200 Available x2 · x1 $1.80/HR 80 GB VRAM · 184 GB RAM · 32 vCP H200 Available x2 · x1 $1.23/HR 80 GB VRAM · 184 GB RAM · 32 vCP H200 Available x2 · x1 $0.47/HR 80 GB VRAM · 184 GB RAM · 32 vCP B300 Available x2 · x1 $4.99/HR 288 GB VRAM · 480 GB RAM · 48 vCP B200 Available x2 · x1 $3.49/hr 192 GB VRAM · 384 GB RAM · 32 vCP H200 Available x2 · x1 $3.14/HR 141 GB VRAM · 182 GB RAM · 44 vCPUs H100 Available x2 · x1 $2.43/HR Spot 0.94/HR 80 GB VRAM · 185 GB RAM · 32 vCPUs GH200 Available x2 · x1 $3.14/HR 96 GB VRAM · 480 GB RAM · 72 vCP RTX Pro 6000 Available x2 · x1 $3.14/HR 96 GB VRAM A100 Available x2 · x1 $3.14/HR 80 GB VRAM A40 Available x2 · x1 $3.14/HR 48 GB VRAM Liquid Reserved Clusters Request large-scale clusters from 50+ providers. Sell-back idle GPUs to our spot market. 1.1 Get quotes from 50+ datacenters within 24 hours One request, parallel bids for options, from H100, H200, to  B200, B300, GB300 NVL72 1.2 Re-sell idle GPUs back to our spot market Resell idle node on our spot market or put on our spot market with no manual ops. Reclaim capacity instantly when you need it 1.3 Direct assistance from our research and infra engineering team Dedicated solutions engineer from cluster bring-up through steady-state GET A QUOTE BOOK A CALL FIG.7 Enter GPU name.. B300 SXM6 x 512 SXM6 3-YEAR RESERVED $5.00 /HR/GPU TOTAL $2,560/hr Reserved cost (3yrs) $67,276,800 Idle hrs resold 6,727,680 hrs Cost of idle capacity $33,638,400 Revenue at $8.00/hr/gpu $53,821,440 Profit on idle capacity +$20,183,040 Research. Our Contributions to the Frontier of Open-Source AI DISCOVER Research Recursive Language Models: the paradigm of 2026 How we plan to manage extremely long contexts READ MORE INTELLECT-3: A 100B+ MoE trained with large-scale RL A 100B+ parameter Mixture-of-Experts model trained on our RL stack. SYNTHETIC-2 Release Four million collaboratively generated reasoning traces. INTELLECT-2 Release The first 32B model trained through globally distributed RL. VIEW ON HUGGING FACE Latest Research. SEE ALL Research JUL 10TH, 2026 verifiers v1: Decomposing Tasksets and Harnesses for Agentic RL & Evaluations Research JUL 05TH, 2026 prime-rl gets an Algorithms layer Research JUN 21ST, 2026 RL at 1T Scale: prime-rl Performance Deep Dive Customer Stories Post-training How Zapier Turned AutomationBench Into a Continuous Agent Improvement Loop Post-training How Ramp Used RL to Beat Frontier Models at Spreadsheet Search We’re Hiring Join Prime Intellect We are seeking the most ambitious developers to join our team — in San Francisco or remotely. Please send us examples of your exceptional work. Join us