INTELLECT-3
Prime Intellect · 2025-11-26 · 106.9B parameters
INTELLECT-3 is Prime Intellect's 106B Mixture-of-Experts reasoning model with 12B active parameters, released in November 2025, post-trained from the GLM-4.5-Air-Base checkpoint via supervised fine-tuning followed by large-scale reinforcement learning on a 512-GPU H200 cluster. It supports a 131K context window and is optimized for math, code, science, and multi-step reasoning. The model weights, training frameworks, and environments are open-sourced under MIT and Apache 2.0 licenses.