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.

Trained in collaboration between Arcee, Datology and Prime Intellect.

https://twitter.com/arcee_ai/status/2016278017572495505

Trinity Architecture

Key design choices:
- Interleaved local + global attention (3:1 pattern)
- Grouped-query + gated attention
- New load-balancing method (SMEBU)
- Depth scaled sandwich norm and QK norm

With extreme sparsity, built for long context and fast inference.

We believe the next breakthrough in long-horizon agents is training models to manage their own context.

Introducing our new research direction on Recursive Language Models.

We are sharing our initial experiments showing the promise of RLMs.

primeintellect.ai/blog/rlm First introduced by @a1zhang in Oct 2025, the RLM has access to its inputs through a variable in a persistent Python REPL.

The model can inspect & transform that variable with code, and pipe parts of it into sub-LLMs with tools without ever loading the potentially huge input data into its context.

Introducing INTELLECT-3: Scaling RL to a 100B+ MoE model on our end-to-end stack

Achieving state-of-the-art performance for its size across math, code and reasoning

Built using the same tools we put in your hands, from environments & evals, RL frameworks, sandboxes & more INTELLECT-3 is a 106B parameter Mixture-of-Experts model trained with both SFT and RL on top of the GLM 4.5 Air Base model.

Both stages, including multiple ablations, were carried out on a 512-GPU H200 cluster over the course of two months.

We're scaling our Open-Source Environments Program

As part of this, we're committing hundreds of thousands of $ in bounties and looking for partners who want to join our mission to accelerate open superintelligence

Join us in building the global hub for environments and evals Over the past 2 months, we've crowdsourced 400+ environments and 80+ verified implementations through our bounties and RL residency across:

+ Autonomous AI Research
+ Browser Automation
+ Theorem Proving
+ Subject-Specific QA
+ Legal/Finance Tasks
+ Many more...

Another week, another hundred environments.

From autonomous AI research, MCP integrations, and browser automation to domain specific environments for economically valuable tasks across law, finance, and tax. NanoGPT Speedrun

Evaluate code-generation and pretraining capabilities of LLMs via NanoGPT Speedrun benchmark.

By @leloykun
app.primeintellect.ai/dashboard/envi…

Today we're launching Reserved Instances

- Request 8–1,000+ GPU clusters
- Get quotes from up to 50+ providers in 24h
- Re-sell idle GPUs back to our spot market
- Support from our research team Expanding our Compute Exchange

- Find the best and most cost-effective reserved instance offers across 50+ providers
- Re-sell idle GPUs from your reserved cluster on our liquid compute market
- H100s, H200s, B200s, and NVL72 clusters available today

Environments Hub launched a week ago, and we’ve already crowdsourced 100+ environments.

Ranging from theorem proving, kernel generation, scientific qa, browser-use, and more. Every environment contributed shifts the balance of power towards open-source AI.

Some highlights: Lean 4 Theorem Proving

Multi-turn formal theorem proving in Lean 4, where models alternate between reasoning, sketching proof code, receiving feedback.

Ideal for search guided rl, process rewards, and curriculum design.

By @myainotez

app.primeintellect.ai/dashboard/envi…

Join the Prime Intellect RL Residency

Our community already shipped 100+ environments to the Environment Hub

Help us accelerate, with compute, a stipend, and support from our research team The RL Residency gives you:
— Compute for experiments
— A stipend
— Hands-on support from our internal research team

Introducing the Environments Hub

RL environments are the key bottleneck to the next wave of AI progress, but big labs are locking them down

We built a community platform for crowdsourcing open environments, so anyone can contribute to open-source AGI Environments are where agents learn.

They define the world, rules, and feedback loop of state → action → reward. Everything from coding/math tasks to games and multi-turn dialogue evals can be thought of as environments. Without them, RL is just math with nothing to interact with.

Launching SYNTHETIC-2: our next-gen open reasoning dataset and planetary-scale synthetic data generation run.

Powered by our P2P inference stack and DeepSeek-R1-0528, it verifies traces for the hardest RL tasks.

Contribute towards AGI via open, permissionless compute. Planetary-Scale Inference
Our peer-to-peer decentralized inference stack moves into production, enabling everyone—from consumer GPUs to hyperscale clusters—to contribute meaningfully towards open-source AI progress.

Releasing INTELLECT-2: We’re open-sourcing the first 32B parameter model trained via globally distributed reinforcement learning:

• Detailed Technical Report
• INTELLECT-2 model checkpoint

primeintellect.ai/blog/intellect… To train a model with reinforcement learning in a fully decentralized setting using community-contributed GPUs, we open-source several novel infrastructure components.

Today we’re launching INTELLECT-2:

The first decentralized 32B-parameter RL training run open to join for anyone with compute — fully permissionless.

Scaling towards frontier reasoning across coding, math and science. INTELLECT-2 brings decentralized training into the inference-time compute era:
• Fully async, decentralized reinforcement learning
• Eliminating communication overhead
• Scalable across heterogeneous GPUs worldwide

primeintellect.ai/blog/intellect…

Today, we're laying the foundation to accelerate open & decentralized AI

Introducing our protocol & testnet: A peer-to-peer compute and intelligence network. Enabling collective creation, ownership, and access of sovereign open-source AI

Towards an open superintelligence future This marks a key next step in our masterplan, leveraging our work on:
- Aggregating compute with our Compute Exchange
- Scaling distributed training with prime, proven in INTELLECT-1
- Enabling distributed reinforcement learning with genesys, battle-tested in SYNTHETIC-1
- Allowing trustless verification of compute through TOPLOC

Our protocol unifies these components and forms the coordination and incentive layer to enable scale.

Introducing SYNTHETIC-1: Collaboratively generating the largest synthetic dataset of verified reasoning traces for math, coding and science using DeepSeek-R1.

Join us to contribute compute towards state-of-the-art open reasoning models. Today, we release:
- SYNTHETIC-1: 1.4 million high-quality tasks & verifiers
- Public synthetic data run - allowing anyone to contribute compute
- GENESYS: open, extendable synthetic data generation framework + call for crowdsourcing tasks & verifiers

primeintellect.ai/blog/synthetic…

Today, we release TOPLOC: A Locality Sensitive Hashing Scheme for Verifiable Inference

- Detects modifications to models, prompts, or precision
- Robust across GPU types, tensor parallel configurations and attention kernels
- Up to 100× faster validation than generation
- Reduces memory overhead of proofs by 1000×

primeintellect.ai/blog/toploc

Building the foundation for decentralized, verifiable compute protocols. The Problem: Trust in LLM Inference

In a peer-to-peer setting, ensuring honest behavior among providers requires detecting and penalizing dishonest ones. Providers often make changes, such as:

- Lowering precision
- Compressing KVCache
- Altering model weights or prompts

Releasing METAGENE-1: In collaboration with researchers from USC, we're open-sourcing a state-of-the-art 7B parameter Metagenomic Foundation Model.

Enabling planetary-scale pathogen detection and reducing the risk of pandemics in the age of exponential biology. METAGENE-1 is a 7B parameter autoregressive transformer model trained on over 1.5T DNA and RNA base pairs sequenced from wastewater samples.

Website: metagene.ai
Paper: metagene.ai/metagene-1-pap…
Github: github.com/metagene-ai/me…
Hugging Face: huggingface.co/metagene-ai

Releasing INTELLECT-1: We’re open-sourcing the first decentralized trained 10B model:

- INTELLECT-1 base model & intermediate checkpoints
- Pre-training dataset
- Post-trained instruct models by @arcee_ai
- PRIME training framework
- Technical paper with all details This represents a 10× scale-up from our previous work and demonstrates that large-scale model training is no longer confined to large corporations but can be achieved through distributed, community-driven approaches.

Technical report: github.com/PrimeIntellect…

Blogpost: primeintellect.ai/blog/intellect…

Introducing Prime Intellect – democratizing AI development at scale, from compute to intelligence.

We're excited to announce our $5.5M raise from @DistributedG @coinfund_io @CompoundVC @Collab_Currency @protocollabs @ClementDelangue @dylan522p and others

primeintellect.ai/blog/introduci… Our vision
Build infrastructure to aggregate compute, develop distributed training frameworks, and create a protocol for decentralized AI development—enabling anyone to contribute resources, collectively train open models, and share in their ownership.