SAN FRANCISCO – Two years ago, a small team of AI researchers published a model called Hermes to GitHub. They did not charge for it. They did not require an API key. Anyone could download it and run their own AI agent on a laptop, a cheap server, or a rented cloud instance. Today, Hermes has more than 214,000 GitHub stars and roughly 40,000 forks. That is not a hobby project. It is a platform, and investors are beginning to treat it as one.
Nous Research, the company behind Hermes, is in active talks to raise at least $75 million in a new funding round that would value the startup at $1.5 billion, TechCrunch reported Sunday. Robot Ventures, which already holds a stake in the company, is expected to lead the round. No deal has closed, and terms could shift before any agreement is signed.
The new raise would more than double what the company has collected in total. Nous Research has previously raised approximately $70 million from Paradigm, Robot Ventures, North Island Ventures, OSS Capital, and individual investor Balaji Srinivasan. The startup was founded in 2023 by Jeffrey Quesnelle, Karan Malhotra, Ryan Teknium, and Shivani Mitra. The company did not respond to requests for comment, and investors declined to participate in the story.
Hermes does the things most AI assistants promise but frequently underdeliver on. It can search the web, write and run code, interpret images, and add new capabilities by auto-learning skills without requiring a software update. That last feature is less common than it sounds. Most deployed agents require a retraining cycle or a plugin system to gain new abilities; Hermes acquires them more fluidly. The model integrates directly with Telegram and Discord, two platforms where developers and independent builders already spend their time.
The model runs on three tiers. A desktop version installs on your own hardware. A VPS deployment costs between $20 and $200 per month depending on compute needs. Cloud hosting falls in the same price range. For a company that built its audience by giving the model away free, the pricing structure represents a deliberate shift toward recurring revenue. It is a common transition in open-source software businesses, and it rarely goes smoothly.
Open-source AI companies commanding billion-dollar valuations are still unusual enough to be worth examining. The business logic is genuinely complicated: the more widely a model is adopted, the harder it becomes to charge for access to that same model. Nous Research’s answer is to sell deployment and infrastructure rather than the model itself. Whether that margin structure can sustain a $1.5 billion exit valuation is a question the funding round has not answered. It has merely placed a bet.
The broader environment for AI fundraising has been unusually aggressive. Among the technology companies making capital markets news this month, General Fusion’s Nasdaq debut illustrated how investors are treating companies at the intersection of emerging technology and long-term infrastructure. Nous Research operates in a different space, in software and model weights rather than physical reactors, but the investor appetite for high-multiple plays on the AI frontier has been consistent.
The AI agent market that Hermes competes in includes paid products from OpenAI, Anthropic, Google, and a rapidly expanding set of enterprise software platforms. Most of those agents run on proprietary models that users cannot inspect or run themselves. Nous Research’s argument, embedded in its open-source approach, is that developers who need control over their AI stack will eventually prefer something they can audit, modify, and self-host. The 214,000 GitHub stars suggest a large segment of that audience already agrees.
The $1.5 billion figure is a term sheet number, not an audited enterprise value. As TechCrunch reported, the talks are ongoing and the round has not closed. Private valuations at this stage carry significant uncertainty, particularly for companies whose core product is distributed freely. Paradigm and Robot Ventures are financial backers with portfolio incentives; their valuation assumptions are not neutral market signals.
What the funding round does not resolve is the core tension in Nous Research’s business. Hermes is genuinely capable and genuinely free. The startup built its reputation by refusing to restrict access, and a large community formed around that decision. It has not yet demonstrated that the same people who downloaded the model will pay for the infrastructure to run it at scale. The proof of that will come after the check clears, not before.
