London's Model ML raises $250 million to automate junior banking work

London-based startup Model ML has raised $250 million in Series B funding at a $1.5 billion valuation. General Atlantic and QED Investors co-led the round, according to TechCrunch, which reported the raise on 17 November 2026.
The company builds AI agents, which are software systems that carry out multi-step tasks without a person directing every step. Its agents build financial models, prepare pitch decks and write due diligence reports. These tasks have long formed the bulk of analyst and associate workloads at investment banks and consultancies.
The agents are designed to sit inside the tools those teams already use. They integrate with clients' systems, data and tools, including Excel and PowerPoint, and can update financial models and presentations automatically when new developments affect the analysis. A revised assumption or a new data point can flow through to the finished deliverable without a manual rebuild.
Model ML was co-founded by brothers Chaz and Arnie Englander. The Series B follows a $75 million Series A announced a year earlier, according to a Model ML press release on PR Newswire and the TechCrunch report.
The company says it runs on open-source models, meaning models whose weights and code are publicly available, and that this allows it to operate at about 2% of the cost of frontier models, the most capable systems offered by the largest AI labs.
In my view, the cost claim will draw the most scrutiny from engineers. A 2% cost ratio is a claim about inference economics, the expense of running a model rather than training it. That figure will depend on workload mix, context length (how much text the model processes at once), retrieval design and how often a document-heavy diligence pipeline is rerun when inputs change. Readers will want the evaluation methodology before treating it as a benchmark.
The harder engineering problem, in my view, is state. An agent writing into a live Excel model or PowerPoint deck has to track provenance, detect broken links and catch formula outputs that look sensible but are wrong. It also has to leave an audit trail that a reviewing associate or a compliance function can follow. These requirements are less glamorous than a funding headline, but they determine whether automated edits survive contact with a deal team.
The broader context here is the apprenticeship model. Investment banking and consulting have historically used junior analyst work, including repetitive modelling and document drafting, as the training ground where judgment is built. If agents take on more of that work, firms will need to rethink how junior staff learn, and the remaining roles may shift toward reviewing, challenging and framing the output.
Investors have priced the company at $1.5 billion. In my view, the practical question for deal teams is narrower: whether faster turnaround on models, decks and diligence reports leaves more time for the judgment calls those documents are meant to support.


