Mika Senghaas

Research and engineering

Senghaas works as a research engineer at Prime Intellect, and the publications listed on Senghaas's site focus on decentralized training and reinforcement learning. Other listed work includes DiLoCo-SWARM, a report on scalable decentralized training.¹

Senghaas is first author of the PRIME-RL report and of SYNTHETIC-2, and a co-author of the INTELLECT-2 report, which describes a reasoning model trained through globally decentralized reinforcement learning. Senghaas is the first named author on the INTELLECT-3 technical report (2025), whose model was trained with prime-rl and environments from the verifiers library.¹⁴²⁵

Work and contributions

• INTELLECT-3 Open weights

  First named author of the INTELLECT-3 technical report (2025), which describes the model's supervised fine-tuning and large-scale reinforcement learning starting from GLM-4.5-Air-Base.²¹

• prime-rl Open-source code

  First author of the PRIME-RL report on asynchronous and decentralized reinforcement learning training at scale, which accompanies the prime-rl repository.¹⁶

• SYNTHETIC-2 Open data

  First author of SYNTHETIC-2, a Prime Intellect project on distributed synthetic data generation for verified reasoning.¹

Timeline

1. 2025First author of the PRIME-RL report on asynchronous and decentralized reinforcement learning training.¹
2. 2025First author of SYNTHETIC-2, a project on distributed synthetic data generation for verified reasoning.¹
3. May 2025Co-authored the INTELLECT-2 technical report.⁴
4. Dec 2025Posted the INTELLECT-3 technical report as first named author.²

Where they work

• Prime Intellect 3 people in the wiki

Related people

• Sami JaghouarAlso at Prime Intellect
• Johannes HagemannAlso at Prime Intellect
• Susan ZhangTraining systems · Language models
• Jakub PachockiLanguage models · Reinforcement learning
• Edward BeechingLanguage models · Reinforcement learning
• Weizhu ChenLanguage models · Training systems
• John SchulmanReinforcement learning · Language models
• Alexander WettigLanguage models · Reinforcement learning
• Tom BrownTraining systems · Language models
• Quoc LeLanguage models · Training systems