ResearchPrime Intellect·25d ago·3 min readSummaryListen 0:51
GLM-5.2 RL weight transfer in 4 seconds using NIXL and ModelExpressGLM-5.2 RL weight transfer completed in 4 seconds using NIXL and ModelExpress.
News Hub
What actually shipped in agent engineering, pulled from the labs, arXiv and Hacker News.
See who we follow34 articles
ResearchPrime Intellect·40d ago·3 min readSummaryListen 0:47
Prime Flash MoE - Faster MoE Kernels optimized for BlackwellPrime Flash MoE provides faster mixture-of-experts kernels optimized for Blackwell GPUs.
ResearchPrime Intellect·46d ago·3 min readSummaryListen 0:38
Multi-Agent Systems in PRIME-RLMulti-agent systems implemented in PRIME-RL framework.
ResearchPrime Intellect·62d ago·3 min read
Scaling Agentic RL: 365,000+ Environments for SWE, Terminal, and SearchPrime Intellect scaled agentic reinforcement learning to 365,000 plus environments for software engineering, terminal, and search tasks.
ResearchPrime Intellect·74d ago·3 min read
verifiers v1: Decomposing Tasksets and Harnesses for Agentic RL & EvaluationsPrime Intellect released verifiers v1 for decomposing tasksets and harnesses in agentic reinforcement learning and evaluations.
ResearchPrime Intellect·110d ago·3 min read
Post-Training Nemotron 3 on LabPrime Intellect enabled post-training of NVIDIA Nemotron 3 models on Lab.
ResearchPrime Intellect·110d ago·3 min read
Prime Intellect Joins the NVIDIA Nemotron Coalition to Advance Open Frontier ModelsPrime Intellect joined NVIDIA's Nemotron Coalition to develop open frontier AI models.
ResearchPrime Intellect·117d ago·3 min read
Releasing Hosted Evaluations: Making benchmarking effortlessPrime Intellect released Hosted Evaluations for simplified benchmarking.
ResearchPrime Intellect·127d ago·3 min read
General Agent: A Self-Evolving, Synthetic Agent EnvironmentGeneral Agent is a self-evolving synthetic agent environment.
ResearchPrime Intellect·133d ago·3 min read
renderers: Token-Level Templating for Agentic RLRenderers provides token-level templating for agentic reinforcement learning.
ResearchPrime Intellect·138d ago·3 min read
Releasing Lab: the training platform for self-improving agentsLab is a training platform for self-improving agents.
IndustryPrime Intellect·159d ago·3 min read
Launching FrontierSWE on the Environments HubFrontierSWE launched on the Environments Hub.
ResearchPrime Intellect·176d ago·3 min read
Partnering with Browserbase to Train Browser and Computer Use AgentsPrime Intellect partners with Browserbase to train agents for browser and computer use tasks.
ResearchPrime Intellect·190d ago·3 min read
Leveraging NVIDIA to Build the Open Superintelligence StackTitle alone insufficient to summarize open superintelligence stack using NVIDIA.
IndustryPrime Intellect·224d ago·3 min read
Introducing Lab: The Full-Stack Platform for Training your Own ModelsPrime Intellect launched Lab, a full-stack platform for training custom models.
ResearchPrime Intellect·264d ago·3 min read
Recursive Language Models: the paradigm of 2026Recursive language models represent an emerging paradigm for 2026.
ResearchPrime Intellect·300d ago·3 min read
INTELLECT-3: A 100B+ MoE trained with large-scale RLINTELLECT-3 is a 100B+ parameter mixture-of-experts model trained with large-scale reinforcement learning.
Open sourcePrime Intellect·330d ago·3 min read
Scaling Our Open-Source Environments ProgramPrime Intellect expands its open-source environments program.
IndustryPrime Intellect·439d ago·3 min read
SYNTHETIC-2 Release: Four Million Collaboratively Generated Reasoning TracesPrime Intellect released SYNTHETIC-2 with four million collaboratively generated reasoning traces.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.