Skip to content
← Models
Open weightsPrime Intellect/10 trustVerified

INTELLECT-2

The first 32B model trained by fully decentralized async RL — a volunteer GPU swarm, Apache 2.0, beating QwQ-32B.

Visit official page →

About

INTELLECT-2 is a milestone in how models get built, not just how smart they are. Prime Intellect trained this 32B Qwen2-based reasoner with fully asynchronous reinforcement learning across a permissionless swarm of volunteer GPUs scattered over the public internet — no datacenter, no InfiniBand, just new infrastructure (PRIME-RL, SHARDCAST weight broadcasting, TOPLOC validators to keep untrusted workers honest) holding the run together.

It worked: the model beats QwQ-32B on key reasoning benchmarks per the team's report, using rule-based rewards on math and coding problems. Everything is Apache 2.0 — weights, code, training logs — which makes INTELLECT-2 both a capable reasoner and the reference implementation for anyone curious whether frontier training really needs Big Tech's clusters.

Specs

Context window

Not officially published (Qwen2-32B base)

Pricing

Free / open weights

Free quota

Fully open — no API

API access

No official API — self-host from Hugging Face (PrimeIntellect/INTELLECT-2)

Weights

Open weights

Released

Not disclosed

Benchmarks

  • Vendor-reported: exceeds QwQ-32B on key reasoning benchmarks (no precise headline numbers independently verified)
  • Technical report: arXiv 2505.07291 on decentralized async RL (PRIME-RL, SHARDCAST, TOPLOC)

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

mathematical-reasoningdecentralized-trainingrl-researchcode-reasoning

Trust score — 8/10

/10 trustVerified

Scored 8/10 (TIER verified, metrics-v1). Evidence: Prime Intellect release (May 2025) with technical report arXiv 2505.07291; Apache-2.0 license confirmed via HF API; open weights, code and training logs; covered by Wikipedia and independent press. Strengths: landmark decentralized-training achievement, fully permissive license, transparent infrastructure. Docked for: (1) benchmark claims are vendor-reported without precise headline numbers; (2) niche adoption vs centralized-lab models. License: Apache-2.0.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
7.0/10
Maintenance · recency, releases, issue load
8.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · Prime Intellect; released 2025-05-11
  • · First 32B model via decentralized async RL; arXiv 2505.07291
  • · Benchmarks are vendor-reported, no precise headline numbers

Sources