WillItSurvive
No grouping · United States

Nous Research

nousresearch.com ↗

Maker of open-weight Hermes AI models and the Hermes Agent, plus decentralized training tool Psyche.

Losing its protection
Assessed 30 July 2026 · Evidence as of 30 July 2026
The structural protection is eroding.
On the recordSince 30 July 2026 · Checkable from Q3 2026 · Every call on the record Checkpoint: Confirmation or collapse of the reported $75M funding round at a $1.5B valuation
The call

Nous has real developer traction but sits in the most crowded, fastest-commoditizing corner of AI, with platform gatekeepers and bigger open-model makers both able to squeeze it.

Nous Research builds free AI models (like Hermes) that lots of developers download, plus software to run AI agents, and research into decentralized AI training called Psyche. The problem is that Meta, OpenAI, Anthropic, and DeepSeek all make similar free models that work just as well or better, and they have way more money and engineering power to improve them. Anthropic has already blocked people from using Nous's agent software through their paid services, showing these big companies can simply lock competitors out of their platforms. The one thing that could save Nous is if Psyche becomes a real product that companies actually pay for and need, rather than staying as just research, and if the company can actually close its funding and be clear about what kind of investment it is selling.

Why
01
AI pressure: Meta Llama, DeepSeek, Alibaba Qwen, OpenAI gpt-oss open-weight models. Rival labs ship free open-weight models that match or beat Hermes benchmarks, squeezing its pricing and download edge. Could AI do the core job? Yes, likely.
02
Bigger players: Anthropic. Anthropic blocked third-party harnesses like Hermes Agent from subscription quotas in April 2026, cutting into agent economics. Could a giant copy it? Yes, easily.
03
What protects it. A large, active open-source developer base: 33M+ Hermes downloads, 218K GitHub stars, and a #1 OpenRouter ranking keep usage flowing to Nous.
What would change this call
Whether the reported $75M/$1.5B round closes with an SEC Form D or S-1 filing. Confirms real capital backing versus an unclosed, unverified valuation claim.
Whether Psyche signs a named paying enterprise customer for decentralized training. Shows the infrastructure moat is monetized, not just a research demo.
Whether OpenAI or Google follow Anthropic in restricting third-party agent harnesses. Would show platform gatekeeping is spreading and squeezing Hermes Agent's economics.
Key numbers
  • Hermes Agent hit #1 on OpenRouter, 224B tokens/day (2026-05-10)
  • 218K GitHub stars, 28.8T cumulative tokens processed (2026-07-21)
  • $70M raised to date across seed and Series A (2026-07-13)
Named risks
  • Core revenue model (inference, Portal subscriptions) exposed to industry inference cost decline of about 92 percent over three years. Severity: high
  • Platform dependency on Anthropic and other model providers who can block or throttle third-party agent harnesses like Hermes Agent. Severity: high
  • Reported 1.5 billion dollar valuation and 75 million dollar round unconfirmed by SEC filings as of July 29, 2026. Severity: medium
  • Potential NOUS ecosystem token launch tied to Psyche could trigger securities regulatory scrutiny, given SEC Crypto Task Force mention in September 2025. Severity: medium
  • Hermes model benchmarks trail closed frontier models like gpt-oss-120b, weakening differentiation versus incumbents. Severity: medium
Rivals, as read in July 2026
RivalDirectionEvidence
OpenClawcontestingOpenClaw has 370K-384K GitHub stars vs Hermes Agent's 218K as of July 2026, though Hermes topped OpenRouter rankings May 10, 2026
Anthropiclosing toAnthropic blocked Hermes Agent from Claude subscription quotas in April 2026, raising Nous's user cost structure
Meta (Llama)contestingMeta ships open-weight Llama models at greater scale and compute than Nous's Hermes family
OpenAI (gpt-oss-120b)losing toHermes 4 70B benchmarked below gpt-oss-120b on Artificial Analysis, published August 24, 2025
DeepSeekcontestingDeepSeek's open-weight models compete directly against Hermes in the open-source model space as of 2026
Gensparklosing toGenspark surpassed $200M annualized revenue with a $385M Series B in April 2026, dwarfing Nous's estimated $9.9M revenue
Questions people ask
Is Nous Research profitable in 2026?
No public profitability data exists. Data aggregators ZoomInfo and Datanyze estimate Nous Research revenue at roughly 9.9 million dollars as of mid 2026, far below its reported 1.5 billion dollar valuation being discussed in July 2026, with no audited financials disclosed.
Who are Nous Research's biggest competitors?
OpenClaw is Nous Research's closest open-source agent rival, with roughly 370,000 to 384,000 GitHub stars versus Hermes Agent's 218,000 as of July 2026. Frontier labs Meta, DeepSeek, OpenAI and Anthropic also ship competing open and closed models.
Could AI or bigger labs replace Nous Research's Hermes models?
Yes, exposure is high. Hermes 4 70B benchmarked below OpenAI's gpt-oss-120b on Artificial Analysis as of August 24, 2025, and industry inference costs fell about 92 percent over three years, per a February 2026 report, pressuring Nous's model and agent pricing.
Who owns Nous Research?
Nous Research is privately held and VC-backed, with founders including Jeffrey Quesnelle, Karan Malhotra, Ryan Teknium and Shivani Mitra. Investors include Paradigm, Robot Ventures, USV, North Island Ventures, Delphi Ventures, OSS Capital and angel Balaji Srinivasan as of July 2026.
What is Nous Research's valuation in 2026?
TechCrunch reported July 13, 2026 that Nous Research was finalizing a 75 million dollar round led by Robot Ventures with USV at a 1.5 billion dollar valuation, up from an April 2025 1 billion dollar token valuation, though no SEC filing confirmed it by July 29, 2026.
Does Anthropic threaten Nous Research's Hermes Agent business?
Yes. In April 2026 Anthropic blocked third-party agent harnesses including Hermes and OpenClaw from Claude Pro and Max subscription quotas, forcing pay-as-you-go billing, though a detection bug was later found and refunds were issued by April 30, 2026.
Timeline
  • 2026-07-21Hermes Agent reaches about 218,000 GitHub stars and 28.8 trillion cumulative tokens processed
  • 2026-07-13TechCrunch reports Nous Research finalizing 75 million dollar round at 1.5 billion dollar valuation
  • 2026-04-30Anthropic's harness-detection bug and Claude Max quota block on Hermes and OpenClaw acknowledged, refunds issued
  • 2026-02-25Hermes Agent officially released
  • 2025-12-03Hermes 4.3 (36B) released, first production model post-trained entirely on Psyche
  • 2025-05Consilience, a 40B parameter model trained on 20T tokens via Psyche, launches
  • 2025-04-25Paradigm leads 50 million dollar Series A at 1 billion dollar token valuation
  • 2025-01-27Psyche Network launches on Solana
  • 2024-08DisTrO distributed-training optimizer published
  • 2024-01-10Seed round of about 5.2 million dollars closes, co-led by Distributed Global and OSS Capital
Revision history
30 July 2026
First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.
In the same band
American Airlines GroupSafe Superintelligence Inc.Wayfair
More companies losing its protection →
Sources