News
Curated articles, my blog posts, and site announcements.
All Categories
Loading news...
Prime Intellect News
Company-specific announcements, X posts, and blog updates from the portfolio.
Portfolio news
43
Exo7 news items
Lighter13 news items
Lucis5 news items
MegaETH14 news items
Monad19 news items
Morpho16 news itemsPrime Intellect39 news items
Accountable8 news items
Dria7 news items
Edison4 news items
Friend4 news items
Octet2 news items
Runner7 news items
Agora5 news items
Aligned Layer17 news items
Delta2 news items
Fabric Cryptography8 news items
Giza17 news items
Praxis5 news items
Rhinestone9 news items
Sorella10 news items
Succinct21 news items
Wildcat9 news items
Zenith7 news items
OWN6 news items
Phylax26 news items
Berachain12 news items
Clique10 news items
Herodotus8 news items
Inco25 news items
Intuition6 news items
Pimlico15 news items
Ritual34 news items
Confetti1 news items
Eclipse18 news items
Knox14 news items
Mizu6 news items
Mode9 news items
Movement13 news items
Nillion5 news items
OpenQ3 news items
PIN AI12 news items
Pragma16 news itemsSort:
- AI
Prime Intellect launches Hosted Evaluations
Prime Intellect launched Hosted Evaluations on its platform: pick an environment and a model, and the platform runs the harness, sandboxes, and parallel compute end-to-end. The Environment Hub hosts hundreds of public leaderboards (GPQA, HLE, SWE-bench Pro, and more) with fully transparent traces and parameters, accessible through the web UI or the prime CLI.
- BlogResearch
Systematic Reward Hacking and Prime Sprints
Prime Intellect publishes research on systematic reward hacking in RL and launches Prime Sprints, a community compute program for studying hack frequency under controlled conditions.
- X Post
Introducing General-Agent: a self-evolving synthetic RL environment
Prime Intellect unveils General-Agent, a fully synthetic environment whose task corpus self-evolves and grows harder over time — 4,504 tool-use tasks across 1,040 domains and 8,159 unique tools, generated by a synthesizer/solver 2-player loop.
- BlogAI
General-Agent: A Self-Evolving Synthetic Agent Environment
Prime Intellect introduces General-Agent, a fully synthetic environment with 4,504 tool-use tasks across 1,040 domains and 8,159 unique tools that self-evolves and grows harder over time.
- AI
Claude Code and Codex autonomously beat the nanoGPT speedrun human baseline
Prime Intellect ran Claude Code (Opus 4.7) and Codex (GPT 5.5) autonomously on the nanoGPT speedrun optimizer track using idle compute — ~10k runs across ~14k H200 hours. Opus now holds the record at 2930 steps vs the 2990 human baseline.
- BlogResearch
Autonomous AI research for the nanoGPT speedrun
Prime Intellect runs Claude Opus 4.7 and GPT 5.5 autonomously on the nanoGPT speedrun, achieving a new 2930-step record vs the 2990 human baseline across ~10k runs and 14k H200 hours.
- AI
Prime Intellect partners with LangChain on self-improving agents
Prime Intellect partners with LangChain to deploy self-improving agents that learn continually from production environments.
- BlogResearch
Renderers: Token-Level Templating for Agentic RL
Prime Intellect releases Renderers, a token-level templating system that fixes chat-template mismatches between RL trainers and environments and unlocks more than 3x throughput on popular open models.
- BlogProduct
Lab opens publicly as the training platform for self-improving agents
Prime Intellect's Lab exits beta, opening public access to a training platform aimed at building self-improving agents from production data.
- AI
Prime Intellect Lab exits beta
Prime Intellect Lab is out of beta, opening access for users to train their own models for self-improving agents.
- BlogProduct
FrontierSWE launches on the Environments Hub
Prime Intellect adds FrontierSWE, a software-engineering evaluation environment, to the Environments Hub.
- BlogAI
Prime Intellect partners with Browserbase to train browser and computer-use agents
Prime Intellect partners with Browserbase to enable training of browser and computer-use agents inside the Environments Hub.
- AI
Leveraging NVIDIA to Build the Open Superintelligence Stack
Prime Intellect outlines how its collaboration with NVIDIA supports an open superintelligence stack spanning Blackwell clusters, Vera CPUs, and Dynamo-powered inference for long-running agentic models.
- BlogInfrastructure
Prime Intellect collaborates with NVIDIA on the Open Superintelligence Stack
Prime Intellect announces a collaboration with NVIDIA to build out the open superintelligence infrastructure stack.
- X Post
TBPN: Prime Intellect’s Will Brown on continual learning in 2026
Will Brown says continual learning could be solved in the first half of 2026, framing it as primarily an engineering problem.
- X Post
Introducing Lab: A full-stack platform for training your own agentic models
Introducing Lab: A full-stack platform for training your own agentic models Build, evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab.
- BlogProduct
Introducing Lab: a full-stack platform for training your own models
Prime Intellect unveils Lab, a full-stack training platform letting users train custom models end-to-end on its compute network.
- X Post
Introducing Trinity Large: 400B MoE model
We're excited to introduce @arcee_ai's Trinity Large model. An open 400B parameter Mixture of Experts model, delivering frontier-level performance with only 13B active parameters.
- BlogResearch
Recursive Language Models: the paradigm of 2026
Prime Intellect proposes Recursive Language Models as the defining architectural paradigm for 2026.
- BlogAI
INTELLECT-3: a 100B+ MoE trained with large-scale RL
Prime Intellect releases INTELLECT-3, a 100B+ parameter Mixture-of-Experts model trained with large-scale reinforcement learning.
- BlogAI
Scaling the open-source Environments program
Prime Intellect expands its open-source environments and bounties program to scale RL environment coverage.
- BlogProduct
Environments Hub launches
Prime Intellect launches the Environments Hub, a community marketplace for reinforcement-learning environments aimed at open AGI.
- BlogAI
SYNTHETIC-2 release: four million collaboratively generated reasoning traces
Prime Intellect releases SYNTHETIC-2, a dataset of four million crowd-generated reasoning traces produced through distributed contributors.
- BlogAI
SYNTHETIC-2 announcement
Prime Intellect introduces SYNTHETIC-2, a distributed project to generate verified reasoning traces at scale.
- BlogInfrastructure
PCCL: Prime Collective Communications Library
Prime Intellect debuts PCCL, a collective communications library purpose-built for distributed AI training workloads.
- BlogAI
INTELLECT-2 release: first 32B model trained through globally distributed RL
Prime Intellect releases INTELLECT-2, the first 32B-parameter model trained end-to-end via globally distributed reinforcement learning.
- BlogInfrastructure
Planetary-scale inference engine for the public internet
Prime Intellect announces a distributed inference engine designed to run model inference across nodes on the public internet.
- BlogAI
INTELLECT-2: kicking off globally distributed RL training of a 32B model
Prime Intellect launches the globally distributed RL training run for INTELLECT-2, its first 32B-parameter model effort.
- BlogFunding
Prime Intellect raises $15M to build the Open Superintelligence Stack
Prime Intellect closes a $15M round to fund the open superintelligence stack spanning compute, training, and inference infrastructure.
- BlogAI
SYNTHETIC-1 release: two million reasoning traces from DeepSeek-R1
Prime Intellect releases SYNTHETIC-1, a public dataset of two million reasoning traces collaboratively generated from DeepSeek-R1.
- BlogAI
SYNTHETIC-1: scaling distributed synthetic data for verified reasoning
Prime Intellect introduces SYNTHETIC-1, a distributed pipeline for generating verified reasoning data at scale.
- BlogResearch
TOPLOC: locality-sensitive hashing for trustless verifiable inference
Prime Intellect presents TOPLOC, a locality-sensitive hashing scheme that enables trustless verification of model inference.
- BlogResearch
Distributed training in the inference-time-compute paradigm
Prime Intellect explores distributed training applied to inference-time-compute reasoning models in the INTELLECT-MATH effort.
- BlogAI
METAGENE-1: metagenomic foundation model
Prime Intellect launches METAGENE-1, a foundation model trained on metagenomic sequence data.
- BlogAI
INTELLECT-1 release: first globally trained 10B parameter model
Prime Intellect releases INTELLECT-1, the first 10B-parameter LLM trained across globally distributed compute nodes.
- BlogAI
INTELLECT-1: launching the first globally distributed training of a 10B model
Prime Intellect kicks off the inaugural globally distributed training run for a 10B-parameter model.
- BlogResearch
OpenDiLoCo: open-source framework for globally distributed low-communication training
Prime Intellect open-sources OpenDiLoCo, a low-communication framework for distributed training across geographically separated nodes.
- BlogProduct
Prime Intellect Compute marketplace launches
Prime Intellect launches its compute exchange, a marketplace for aggregated GPU compute across providers.
- BlogResearch
State of the art in globally distributed training
Prime Intellect publishes its initial overview of distributed training benchmarks and methodology.