Nous Research raises $90M Series B at $1.5B valuation for Hermes Agent

Nous Research, the lab behind the open-source Hermes Agent, confirmed a $90 million Series B on October 7 and is using it to sell that agent to enterprises. The Wall Street Journal reported the valuation at $1.5 billion.

Key facts

• The round is $90 million, led by Robot Ventures, and brings Nous Research's total funding to $158 million, per TechCrunch.
• The valuation is $1.5 billion, per The Wall Street Journal's reporting on the deal — the company's own fundraise note does not state a valuation figure.
• The new money funds Hermes for Businesses, a paid enterprise version of the agent; no price, feature list, or launch date has been published.
• Nous says Hermes Agent has been cloned more than 24 million times and, "to our best internal estimates," drives about 2.5 percent of global token usage — both company self-reports, not independently verified.
• Nous was at roughly $36 million in annualized revenue by mid-September 2026, The Wall Street Journal reported, and expects to pass $100 million before the end of 2026.

What happened

The round was led by Robot Ventures with participation from Nvidia, Union Square Ventures, Menlo Ventures, Samsung, Y Combinator, and 1789 Capital, according to TechCrunch and the company's own fundraise note — which lists Microsoft's M12 among investors as well. TechCrunch counts the three-year-old startup's total funding at $158 million.

Two products anchor the announcement. The first is Hermes Agent itself: the open-source runtime Nous released under the MIT license in February 2026, which users run on their own machines and point at whichever models they choose. The second is new. "With our fundraising and new partners, we will build and deliver Hermes for Businesses," CEO Dillon Rolnick writes in the company's note, framing it as a way for companies to "own their intelligence stacks" rather than rent capability from a hosted API. The note names no price, feature set, or ship date.

The traction numbers attached to the launch deserve care. Nous states that Hermes "has been cloned over 24 million times" and that, "to our best internal estimates," it drives "approximately 2.5% of all global token usage." Both figures come from the company's own blog post. They are internal estimates — not numbers from a router dashboard, an app-store audit, or any third-party measurement — and they should be treated the way we treat every provider self-report at DeAI: as claims awaiting verification, with third-party router dashboards and audited download counts the obvious checks.

The revenue picture rests on WSJ reporting: roughly $36 million annualized by mid-September, with an expectation to cross $100 million before the end of 2026. That trajectory is aggressive for an agent that has, until now, had no paid product surface — the money, in effect, is a bet that enterprises will pay for something they have so far used for free.

Why it matters

For builders, the signal is that an open-weights-native lab has made ownership-of-the-stack its enterprise pitch. Nous's note argues the opposite of the walled-garden model: the agent runs where you run it, takes whatever models you point it at — open weights on your own GPUs, or any hosted or marketplace endpoint — and keeps your data on your side of the line. That pitch pressures hosted-API incumbents on the deployment-control axis, and it lands at a moment when the cheap-fast model tier is repricing downward across the board, as Anthropic's Claude Haiku 5.5 cut this week shows.

The second thing worth watching is what the enterprise product does to the open project. "Hermes for Businesses" is described as open-ecosystem-compatible, but the company has not said how — or whether — the enterprise features (SSO, audit, SLAs, workspace controls) will themselves be open source, per The Decoder's summary of the announcement. How that split resolves will shape whether the enterprise tier strengthens the commons or forks the community's trust in it.

Background

Nous Research was founded in 2023 to push against what its fundraise note calls the pull of centralization in AI. Its early reputation came from post-training work: fine-tuning Meta's Llama models, pioneering synthetic data pipelines, and publishing YaRN, a context-extension method that made open models usable over far longer documents. Hermes Agent, released in February under an MIT license, extended that open posture to the agent layer itself — users choose where it runs and which models it calls, and it carries project context across sessions.

That open-source distribution is what makes the enterprise bet unusual. Most agent startups monetize a hosted runtime from day one; Nous monetized nothing for years, accumulated a user base the company now values in the tens of millions of clones, and is only now building a paid tier. Our earlier coverage of the Hermes stack explains the runtime's model-agnostic design in detail: how to run Hermes 4 and uncensored fine-tunes, and how aggregators and marketplaces compete on routing. The raise also lands as the frontier-vs-open gap keeps narrowing — our running comparison of GPT-5.5 against open models tracks exactly the capability parity an enterprise buyer would weigh against the control an open agent offers.

For context on where the money sits relative to the rest of the sector: a $1.5 billion valuation puts Nous in the same tier as the mid-size model labs that went through this cycle a year earlier, and well below the frontier labs, but with a distribution advantage none of them have — a free, self-hostable agent already on millions of machines.

What's next

Watch for three things. First, a product surface for Hermes for Businesses: Nous has published no price, feature list, or launch date, so any of those landing — likely as an enterprise page on the Hermes Agent site — is the next concrete signal. Second, independent verification of the traction numbers: if OpenRouter or a comparable router dashboard shows Hermes-driven traffic consistent with a multi-percent share of global tokens, that would firm up the claim Nous currently self-reports. Third, whether the enterprise tier's controls are themselves open source, which determines whether the commons that made the company valuable stays whole.