Mirendil Series A Raises $1bn, Led by Kleiner Perkins

Mirendil, an artificial intelligence research and development company, has announced a Mirendil Series A of $1 billion, led by Kleiner Perkins and Clearlake Capital. The company published the announcement on 12 November 2026, dated 9:00am Pacific Time. It did not disclose a valuation.

In the release, Mirendil said the round included participation from Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC, and S32, “among other new and existing investors.” Sidley Austin, writing the following day, said it had represented Clearlake Capital Group, L.P. “in its co-lead investment in the US$1 billion financing of Mirendil.”

Neither the company nor Sidley gave a post-money or pre-money valuation. The $1 billion figure in both texts is the amount raised.

Mirendil described itself as building AI tools that are meant to make frontier research and development more accessible to scientists and engineers. Sidley’s 13 November note uses the same framing: the funding, it said, will support “development of AI tools designed to make frontier AI research and development more accessible to scientists and engineers.” The two statements agree on the intended use of proceeds. They do not add a customer list, a revenue figure or a shipping date for a generally available product.

The company’s own claims about how those tools are built are more specific, and they come only from Mirendil. The release said:

One researcher, working with agents, pretrained Mirendil-Base, a competitive coding model, from scratch in 3.5 months with less than $10M in compute.

That sentence is a company claim about headcount, method and compute spend. It is not accompanied, in the 12 November text, by an independent audit, a training log, or a third-party benchmark. Mirendil also said its Fern system “holds up to 10M tokens of context,” and that Lor, another of its tools, is “now in private beta.” Those product descriptions are the company’s, and Lor is not presented as generally available.

The investor list is long and mixed. Kleiner Perkins and Clearlake are named as co-leads. Google appears as a participating investor, alongside specialist firms such as Battery Ventures, Pear VC, M13 and S32. Because the company added “among other new and existing investors,” the published names should not be read as a complete cap table. The announcement also does not say where Mirendil is headquartered.

A $1 billion Series A is a very large announced raise by recent standards for AI developer tools, even in a week that also brought LangChain’s $300 million Series C. That comparison is about disclosed round sizes only. It says nothing about Mirendil’s valuation, which the company and its counsel have not published.

The “one researcher, working with agents” line sits in a wider industry argument about how much human labour a training run still needs. TSN has recently covered other companies’ claims that agents can write substantial engineering artefacts, including AWS’s account of a local API emulator written by autonomous agents. Those are separate company claims, not independent confirmation of Mirendil’s training story. For a sense of how other labs are packaging agent work for operators, rather than for a single pretraining run, see TSN’s report on self-hosted Claude Managed Agents.

A large raise, and a still-private product

The facts that can be nailed down are narrow. Two primary sources, one from the company and one from counsel to a co-lead, agree that Mirendil raised $1 billion in a Series A co-led by Kleiner Perkins and Clearlake Capital. They agree, at a high level, on the intended use of proceeds. They do not agree on, because they do not state, a valuation.

That omission is worth keeping in view. A $1 billion raise can sit at many different prices. Readers who want a multiple will not find one in these documents, and TSN is not inferring one from the lead firms, from Google’s participation, or from other AI funding announcements in the same week.

The technical claims should be kept in the same register. “Less than $10 million in compute” for a “competitive coding model” would, if independently verified, be a striking training cost. It has not been independently verified here. Fern’s “up to 10 million tokens” is a context-window ceiling, not a measured quality result. Lor’s private beta means early access is gated, so the flagship product named in the announcement is not yet something a customer can simply buy.

Google’s presence on the investor list is notable only as a disclosed participation. It does not, on the published record, describe a product partnership or a cloud commitment, and it should not be stretched into one.

The larger question the raise poses is whether a lab can sell a workflow in which a small number of researchers direct agents through pretraining, evaluation and productisation, and whether customers will trust that workflow while a named tool is still in private beta. The 12 and 13 November texts announce the money and the ambition. They do not yet show a generally available product, a valuation, or an independent test of the training claim.