PrimeIntellect has released INTELLECT-3.1, a 106 billion parameter (A12B) Mixture-of-Experts (MoE) reasoning model.

Model Details

INTELLECT-3.1 is the result of continued training of the INTELLECT-3 model, with the addition of reinforcement learning focused on mathematics, programming, software engineering, and agentic tasks. Training was performed using prime-rl and environments built with the verifiers library. All training and evaluation environments are available on the Environments Hub.

Open Source

The model, training frameworks, and environments are released under fully-permissive open source licenses (MIT and Apache 2.0). Further details are available in the technical report.