Latest Updates fromPrime Intellect.

Rewriting Prime Agent in Rust

On the Nature of the Swarm

Prime Inference: Fast, Reliable Serving for Frontier Open Models

How Extropic Uses Prime Intellect to Train a Thermodynamic ML Research Agent

Prime Sandboxes: MicroVMs for Agentic RL Training at Scale

How Goodfire Uses Prime Intellect to Catch AI Models Cheating at Scale

GLM-5.2 RL weight transfer in 4 seconds using NIXL and ModelExpress

Uncovering a universal offline sandbox escape

Measuring Autonomous AI Research

Prime Flash MoE - Faster MoE Kernels optimized for Blackwell

How Mixedbread Used Prime-RL to Beat Frontier Models in Deep Search

Multi-Agent Systems in PRIME-RL

Prime Agent: A self-improving RLM agent

Scaling Agentic RL: 365,000+ Environments for SWE, Terminal, and Search

verifiers v1: Decomposing Tasksets and Harnesses for Agentic RL & Evaluations

$130M Series A to Build the Open Superintelligence Stack

prime-rl gets an Algorithms layer

RL at 1T Scale: prime-rl Performance Deep Dive

True Agents Model the World

Post-Training Nemotron 3 on Lab

Prime Intellect Joins the NVIDIA Nemotron Coalition to Advance Open Frontier Models

Releasing Hosted Evaluations: Making benchmarking effortless

Systematic Reward Hacking and Prime Sprints

General Agent: A Self-Evolving, Synthetic Agent Environment

Autonomous AI research for nanogpt speedrun

renderers: Token-Level Templating for Agentic RL

Releasing Lab: the training platform for self-improving agents

How Ramp Used RL to Beat Frontier Models at Spreadsheet Search

How Zapier Turned AutomationBench Into a Continuous Agent Improvement Loop

Launching FrontierSWE on the Environments Hub

Partnering with Browserbase to Train Browser and Computer Use Agents

Leveraging NVIDIA to Build the Open Superintelligence Stack

Introducing Lab: The Full-Stack Platform for Training your Own Models

Recursive Language Models: the paradigm of 2026

INTELLECT-3: A 100B+ MoE trained with large-scale RL

Scaling Our Open-Source Environments Program

Environments Hub: A Community Hub To Scale RL To Open AGI

SYNTHETIC-2 Release: Four Million Collaboratively Generated Reasoning Traces

SYNTHETIC-2

PCCL - Prime Collective Communications Library

INTELLECT-2 Release: The First 32B Parameter Model Trained Through Globally Distributed Reinforcement Learning

Planetary-Scale Inference: Building a Distributed Inference Engine for the Public Internet

INTELLECT-2: The First Globally Distributed Reinforcement Learning Training of a 32B Parameter Model

$15M to Build The Open Superintelligence Stack

SYNTHETIC-1 Release: Two Million Collaboratively Generated Reasoning Traces from Deepseek-R1

SYNTHETIC-1: Scaling Distributed Synthetic Data Generation for Verified Reasoning

TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference

Distributed Training in the Inference-Time-Compute Paradigm

METAGENE-1: Metagenomic Foundation Model

INTELLECT-1 Release The First Globally Trained 10B Parameter Model

INTELLECT–1: Launching the First Globally-Distributed Training of a 10B Parameter Model

OpenDiLoCo: An Open-Source Framework for Globally Distributed Low-Communication Training

Introducing Prime Intellect Compute: The compute exchange

State-of-the-Art in Globally-Distributed Training