what's nous research?
Nous Research is an AI research organization focused on developing open-source, human-centric large language models (LLMs) and advanced simulators. The company emphasizes creating accessible AI technologies that prioritize user alignment and ethical considerations in model training and deployment. In the crypto space, Nous Research is pioneering decentralized AI infrastructure through projects like Psyche, a network built on Solana that enables distributed training of AI models using global idle computing resources. This approach leverages blockchain for secure, permissionless participation, allowing contributors to provide GPU power in exchange for rewards while reducing centralization in AI development. The Psyche testnet facilitates collaborative pretraining runs, such as large-scale LLM experiments over trillions of tokens, utilizing innovations like the Nous DisTrO networking stack to optimize inter-GPU communication. This integration of AI and blockchain aims to democratize access to high-performance computing for open-source AI advancement.
Nous hosts Hermes Agent Jam today; launched Autoreason with SHL0MS on April 12.
Links
x.com/nousresearchHosts Hermes Agent Jam event on 2026-04-14 at 4PM EST in the Nous Discord, featuring team presentations and Q&A.
Launched "Autoreason" in collaboration with SHL0MS on April 12, 2026.
Included in Grayscale's Q2 2026 Assets Under Consideration list for potential future investment products.
Provides MIMO V2 Pro free through Nous Portal and Hermes Agent update until 2026-04-21.
Introduces the Manim skill for Hermes Agent, enabling AI agents to create programmatic animations for mathematical and technical explainers using the engine behind 3Blue1Brown.
Integrated Arcee's Trinity large model into Hermes Agent, available through Nous Portal and OpenRouter as of April 2, 2026.
Reached #1 on OpenRouter and achieved 40,000+ GitHub stars.
Launched Hermes Hackathon sponsored by Delphi Ventures with $11,750 prizes, submissions due 2026-03-16. Surpassed 10,000 GitHub stars as users replaced OpenClaw, and fixed migration tool from OpenClaw to Hermes.
Had DeMo paper accepted to ICLR 2026 conference, scheduled to be presented in Rio in April 2026.
Received recognition as one of the serious decentralized AI infrastructure players alongside NEAR Protocol, PrimeIntellect, and Gensyn in a year-end prediction review by a prominent crypto analyst.
Default API rate limit expanded 4x, team teases bigger announcement of improvements coming soon.
Hermes4 model integrates with comput3ai to power interactive experiences for ArtDeCC0s Codex and Museum of Crypto projects, with MVP planned for open source release.
First Hermes model trained entirely with Distro on Psyche (Hermes 4.3 on ByteDance Seed 36B) reaches top spot on RefusalBench leaderboard.
Released Hermes 4.3 AI model, post-trained on Psyche Network on Solana, achieving same performance as Hermes 4 70B at half the size.
Listed among projects expected to have a Token Generation Event in 2026.
Nous Chat now offers free, anonymous access without requiring an account, and the Nous API supports USDC payments on Solana through Coinbase Developer's x402 integration.
@max_paperclips from Nous Research participates as a speaker at dAGI Summit next week.
Psyche Network launches on Solana. Nous Con conference takes place next week in San Francisco.
Released v0.7.0 of Hermes Agent, introducing a self-evolution loop for reusable skill writing and retrieval from memory; hires for agent development and post-training LLMs.
The Hermes API has doubled in tokens generated and requests over the past week.
Training run appears to have been paused or stopped, with speculation that it may not resume operations.
Serious fraud allegations emerge claiming cofounders Karan and Cindy are "fake" and that there is no actual distributed training product, with critics calling the entire distributed training run illegitimate and warning of potential token exit schemes.
Research reveals open source models require 2-3x more tokens than closed source, potentially increasing costs and response times.
8B competitor model outperforms 40B model while raising concerns about closed network structure and whitelist restrictions in decentralized training claims.
Launches Psyche, a decentralized network for training open-source AI models
Successfully completed pre-training of 15B parameter model using distributed computing, marking significant milestone before advancing to 40B model development.
Active funding available with expedited due diligence for ecosystem builders.
$500k compute pool for 40B parameter AI model fills in 44 minutes; launches decentralized training testnet on Solana.
Integrated Tinker with the Atropos async RL framework, enabling training in decoupled environments while leveraging Tinker's managed infrastructure.
SF hackathon scheduled for May 18th featuring $50k prizes and partnerships with NVIDIA and 7 other tech companies.
$50M Paradigm-led funding round secured; Psyche Network testnet goes live.
Paradigm makes strategic investment in AI research project.
Recent funding raise announced amidst growing momentum in decentralized AI training sector.
$50M raised from Paradigm to develop decentralized AI on blockchain.
$50M raised from Paradigm at $1B FDV, with active AI model development and early access program launch.
Decentralized AI lab secures new funding round with institutional backing.
Paradigm leads $50M Series A funding round, valuing decentralized AI infrastructure at $1B.
Secures $50M Series A funding from Paradigm at $1B valuation, utilizing Solana for distributed AI model training.
Critical technical documentation gaps exist around model training, compute resources, and reward mechanisms, with limited transparency compared to similar projects in the space.