INTELLECT-3: Technical Report
Prime Intellect Team, Mika Senghaas, Fares Obeid, Sami Jaghouar, William Brown, Jack Min Ong, Daniel Auras, Matej Sirovatka, Jannik Straube, Andrew Baker, Sebastian Müller, Justus Mattern, Manveer Basra, Aiman Ismail, Dominik Scherm, Cooper Miller, Ameen Patel, Simon Kirsten, Mario Sieg, Christian Reetz, Kemal Erdem, Vincent Weisser
2025-12-24
Summary
This paper introduces INTELLECT-3, a new artificial intelligence model that's really good at tasks requiring thinking, like math, coding, and science. It's a large model, but cleverly designed to be efficient, and the researchers are sharing both the model itself and all the tools they used to build it with the public.
What's the problem?
Creating AI that can truly reason and solve complex problems is incredibly difficult. Existing large language models often require massive amounts of computing power and data, making them expensive and inaccessible. Furthermore, effectively training these models using reinforcement learning – where the AI learns through trial and error – at a large scale presents significant technical challenges.
What's the solution?
The researchers built INTELLECT-3 using a 'Mixture-of-Experts' approach, meaning it only activates parts of the model needed for a specific task, making it smaller and faster than a single, massive model. They trained it using reinforcement learning on a new system called 'prime-rl' which allows for efficient training across many computers. They started with an existing base model (GLM-4.5-Air-Base) and then improved it with both standard training and reinforcement learning, using a collection of environments and tools they also made available.
Why it matters?
This work is important because it demonstrates that it's possible to create a high-performing AI model without needing enormous resources. By open-sourcing both the model and the tools used to create it, the researchers are enabling other scientists and developers to build upon their work, potentially accelerating progress in the field of artificial intelligence and making advanced AI technology more widely available.
Abstract
We present INTELLECT-3, a 106B-parameter Mixture-of-Experts model (12B active) trained with large-scale reinforcement learning on our end-to-end RL infrastructure stack. INTELLECT-3 achieves state of the art performance for its size across math, code, science and reasoning benchmarks, outperforming many larger frontier models. We open-source the model together with the full infrastructure stack used to create it, including RL frameworks, complete recipe, and a wide collection of environments, built with the verifiers library, for training and evaluation from our Environments Hub community platform. Built for this effort, we introduce prime-rl, an open framework for large-scale asynchronous reinforcement learning, which scales seamlessly from a single node to thousands of GPUs, and is tailored for agentic RL with first-class support for multi-turn interactions and tool use. Using this stack, we run both SFT and RL training on top of the GLM-4.5-Air-Base model, scaling RL training up to 512 H200s with high training efficiency.