Launching SYNTHETIC-2: our next-gen open reasoning dataset and planetary-scale synthetic data generation run. Powered by our P2P inference stack and DeepSeek-R1-0528, it verifies traces for the hardest RL tasks. Contribute towards AGI via open, permissionless compute.

Planetary-Scale Inference Our peer-to-peer decentralized inference stack moves into production, enabling everyone—from consumer GPUs to hyperscale clusters—to contribute meaningfully towards open-source AI progress.

Pipeline Parallelism No single GPU holds the full model - each handles a stage, streaming activations forward. This lets smaller GPUs run large models like DeepSeek-R1. Hidden states pass stage to stage; the final GPU decodes a token, sends it back, and the cycle continues.

Verification with TOPLOC v2 To trust results from thousands of permissionless compute nodes, we must verify their generations cheaply. Our TOPLOC verifiable inference work employs a compact locality sensitive hashing scheme for intermediate activations, which can detect unauthorized modifications to models, prompts, or precision. TOPLOC v2 extends this scheme to the pipeline parallel inference setting: • Group-level reward: If the final output is correct, we treat it as evidence that all pipeline stages behaved honestly. • Blame assignment on failure: If a result fails verification, we replay proofs stage-by-stage, pinpoint the first faulty worker, reject the output, and slash that node’s reward.

Protocol We’ve updated our protocol codebase with the following feature and scalability improvements over the past couple of months: • Dynamic Grouping: Worker GPU nodes are now dynamically grouped based on task requirements and geolocation to maximize pipeline utilization and inference and training efficiency. • Extensible Plugin System: A modular plugin architecture now enables reusable components across different compute pools. • P2P Communication Layer: We’ve introduced secure peer-to-peer communication between protocol components - such as the validator and the orchestrator. • AMD Support: ROCm devices such as the MI300X can now join the compute pool. • Observability & Monitoring: We’ve integrated Prometheus dashboards to better monitor the global compute fabric.

SYNTHETIC-2 Dataset A next-gen open dataset for reasoning: • Verified traces from DeepSeek-R1-0528 and Qwen3 for supervised fine-tuning • Difficulty-annotated RL tasks via pass@k from smaller models • 20+ diverse tasks with programmatic verifiers • Includes non-verifiable prompts to broaden SFT coverage beyond math/coding

How to Contribute Compute • Fully Managed: Contribute GPUs directly through Prime Intellect • Self-Hosted: Bring your own idle GPUs Click “Contribute Compute” on the dashboard. Nodes sync, run models, and rank on the leaderboard. Join now:

Live view of our dashboard

Next Steps Building on SYNTHETIC-2, we're launching a new distributed RL run with large-scale async RL building on INTELLECT-2’s success, to train SOTA reasoning models. Coming soon: open-source verifiers, tool-use & coding agents in INTELLECT-3.

Read more in our release