Prime Intellect
Prime Intellect operates an AI training platform and research lab. Its own releases include the INTELLECT series and small models used to study reinforcement learning.
Company & leadership
- CEO
- Vincent Weisser
- CTO
- Johannes Hagemann
- Research lead
- Will Brown
- Legal entity
- PrimeIntellect, Inc.
Founders
- Vincent Weisser and Johannes Hagemann
The founders brought experience in open science and model training
Vincent Weisser co-founded VitaDAO before Prime Intellect. Johannes Hagemann developed language-model training systems at Aleph Alpha. Research lead Will Brown previously worked in Morgan Stanley's machine-learning research group and created the verifiers library for reinforcement-learning environments.
The company raised a Series A in July 2026
Prime Intellect announced a $130 million Series A on 8 July 2026. Radical Ventures led the round. NVIDIA Ventures, Intel Capital, Dell Technologies Capital, and existing investors participated. The company said its total funding then exceeded $150 million.
Prime Intellect acquired Moondream in October 2026
On 13 October 2026, Prime Intellect announced its acquisition of Moondream. Founders Vik Korrapati and Jay Allen joined to lead inference and multimodal work. Prime is integrating Moondream's Photon inference engine. Moondream remains its brand for small multimodal models, and Moondream 3.1 is available through Prime Inference.
Model development and model hosting are different parts of the business
Prime Inference offers serverless access and dedicated deployments. Its catalog includes models from other developers, such as GLM, DeepSeek, and Kimi. These hosted models retain their original developer attribution in this directory.
What this catalog includes
Practical guidance. This profile covers documented Prime-trained INTELLECT releases and small training-research models. Precision packages and intermediate training checkpoints remain with the corresponding model. Third-party mirrors, chat-template changes without new training, random-weight fixtures, and Jacobian-analysis files do not receive separate model pages.
Some public checkpoints have incomplete release notes
The public organization also contains Qwen2.5-0.5B-Reverse-Text-SFT and Qwen3-4B-Instruct-2507-SFT-DeepDive. Their pages lack complete model descriptions. These files remain outside the reviewed model records until their training identity and intended use can be verified.