€3m · Pre-Seed · Legaltech · Munich, Germany
Grubel, a Munich-based AI research lab focused on complex legal work, has raised €3m in pre-seed funding led by Point Nine. The financing also included Jeff Dean, Chris Ré, Ion Stoica, and Harvey co-founders Gabe Pereyra and Winston Weinberg, among other angels. According to EU-Startups, the company will use the capital for research, product development and team expansion.
Founded in 2026 by machine-learning researchers Moritz Hardt and Reinhard Heckel, Grubel is building systems that adapt to an individual legal matter rather than applying the same general model and workflow to every case. Its system combines a data engine that curates matter-specific information, a test-time adaptation layer that adjusts the model and agent, and continual evaluation against the task's own standards.
Specialisation has to become repeatable
Grubel's thesis starts with a practical constraint: complex matters can contain large, unstructured data rooms, while the relevant issues and success criteria differ from case to case. In a company-authored benchmark using synthetic diligence tasks from Harvey's Legal Agent Benchmark, Grubel reported that a specialised harness passed 63.08% of 7,359 criteria across 11 tasks. The tested general-purpose setups scored materially lower, while average model cost for the specialised runs was about $12 per task.
That result is evidence for the technical direction, not yet proof of a commercial product. The benchmark still left roughly 37% of criteria unmet, and Grubel has not publicly disclosed customer names, revenue or production deployments. The pre-seed round therefore funds the step from a promising research result to a repeatable system that legal teams can trust across different matters.
Point Nine says the company plans to reach law firms both through partners such as Harvey and directly. Those routes have different demands. A partner channel could provide distribution and established workflows, while direct sales would require Grubel to carry more of the integration, security and support burden itself.
The potential moat sits in the improvement loop rather than the underlying model alone. Grubel says matter-specific systems can run on a customer's own infrastructure and keep files within its custody. If the company can automate data curation, adaptation and evaluation without rebuilding the process manually for each engagement, it could turn legal teams' private work product into a system-level advantage. The commercial test is whether those gains persist on live matters where accuracy, confidentiality and review obligations are harder than in a synthetic benchmark.
Sources checked:
Grubel company site · Grubel research note · EU-Startups · Point Nine · Tech.eu Funding Explorer
