Nous Research is in talks to raise at least $75 million in new funding at a $1.5 billion valuation, according to three people with knowledge of the deal cited by TechCrunch. Robot Ventures is leading the round, with significant participation from USV and other unnamed investors. Nous Research declined to comment; USV and Robot Ventures did not respond to TechCrunch's request for comment.

The talks come less than three months after Nous Research closed a $50 million Series A, a round led almost entirely by Paradigm at a $1 billion token valuation and built around the company's decentralized AI platform on Solana The Block. Two funding events inside a single quarter, with the valuation moving from $1 billion to $1.5 billion, is a compressed timeline even by the standards of the current AI financing cycle.

Nous Research was founded in 2023 by Jeffrey Quesnelle, Karan Malhotra, Ryan Teknium, and Shivani Mitra. Before this new round, the company had raised a total of $70 million, from Paradigm, Robot Ventures, North Island Ventures, OSS Capital, and Balaji Srinivasan, per Crunchbase data cited by TechCrunch. The new capital, according to TechCrunch's sourcing, will go toward expanding Hermes products and the company's business model — though neither the specifics of that expansion nor a closing date for the round have been disclosed.

Hermes is the open-source agent at the center of Nous Research's pitch to investors. The project has accumulated roughly 214,000 stars and nearly 40,000 forks on GitHub, a scale of community adoption that few open-source agent frameworks have matched. Hermes ships with built-in skills — web search, coding, image understanding — and was designed to learn from usage patterns to build additional skills over time, rather than shipping as a fixed capability set.

Nous Research monetizes Hermes through a cloud-hosted version offered across paid tiers running from $20 to $200 a month, a SaaS-style ladder that sits alongside the open weights rather than replacing them. The company also operates a decentralized compute network, through which users contribute hardware for training and inference workloads — the same decentralization thesis that underpinned the Solana-based platform funded in the Series A.

That dual identity is worth sitting with for a moment. Nous Research is simultaneously an open-source project with genuine grassroots pull on GitHub, a crypto-native infrastructure play raising on token valuations, and now a conventional venture-backed startup courting Robot Ventures and USV at a straight equity-style valuation marker. Few companies run all three plays at once, and each pulls on a different investor base with different expectations for return and governance.

The pace of the raise is the more striking data point here. A jump from a $1 billion token valuation in April to a reported $1.5 billion equity-adjacent valuation in July, without a shipped product milestone disclosed alongside it, reflects how quickly capital is chasing agent infrastructure companies with visible open-source traction right now. Whether that traction — GitHub stars, forks, a hosted tier — translates into durable revenue at a scale that justifies a 50 percent valuation step-up in under three months is the open question any investor in this round is underwriting.

Worth flagging: TechCrunch's report rests on sourcing from three people described as having knowledge of the deal, not a company statement, and both Nous Research and the named investors have so far declined or not responded to comment requests. Deal terms in early-stage and even growth-stage AI rounds have moved fast enough recently that valuations reported mid-negotiation sometimes shift before signing. Readers should treat the $1.5 billion figure as the state of talks as of this reporting, not a confirmed close.

The company's earlier technical work gives some texture to what "expanding Hermes products" might mean in practice. Nous Research's blog previously detailed tinker-atropos, an integration layer connecting the Tinker API with its Atropos reinforcement learning framework — the kind of infrastructure that would let Hermes agents be fine-tuned against reward signals gathered from real usage, consistent with the stated design goal of the agent building new skills autonomously.

For an industry that has watched agent startups proliferate largely on the strength of demo videos and benchmark claims, Nous Research's GitHub numbers offer something closer to load-bearing evidence of actual developer adoption. Whether that adoption converts into the kind of recurring revenue that supports a $1.5 billion valuation, and whether the decentralized compute network scales alongside the hosted product, are the questions this round is effectively betting will resolve favorably. The agent tooling market has enough entrants now that distribution and community trust, not just model quality, increasingly decide who raises the next round at a step-up.