Skip to content
Antenna

Robotics · San Francisco, CA, USA · Founded 2026 · Raised $10M · Formation stage

Robocurve

Tests how well AI models control real robots and publishes the results

Edge

37#10 · Unchanged in 7d

3 of 8 families

Brief

Written by a research agent on Oct 2 from the sources below. A second agent checked each claim.

Robocurve, in San Francisco per the YC page, tests how well AI models control real robots and publishes the results. It describes itself as an independent third-party evaluator and says it is incorporated as a Public Benefit Corporation. It maintains Inspect Robots, an MIT-licensed evaluation framework that lists packages for several commercial robot platforms.

Why it fits

If general-purpose robot policies keep improving, buyers, labs and regulators will need third-party measurement on real hardware. Robocurve is an early attempt to be that reference. It could give investors comparable performance data on models portfolio companies depend on.

Stage and round

Seed stage. The company blog announced a $10M seed round on 14 September 2026, led by Initialized Capital with Notable Capital, Decasonic, Y Combinator, Halcyon Futures and "many others". No valuation is given. An EDGAR full-text search on 2 October 2026 returned nothing for Robocurve. YC lists the Summer 2026 batch. The seed post dates incorporation to about three months before 14 September.

Ask on the first call

  1. 01The pages opened disclose no price, contract or paying customer. What is revenue to date and who are the first paying counterparties?
  2. 02When and why did Aris Zhu, presented as co-founder in July 2026, leave? What happened to the founder equity, and who now leads robotics hardware?
  3. 03How does the charter define independence if evaluated labs are also customers, and do OpenAI or Anthropic have any role in reports on subdomains named after them?
  4. 04Inspect Robots is MIT-licensed. What stops a lab, a university consortium or an existing evaluator such as METR from running the same benchmarks, and what is proprietary?

Risks

  • –No disclosed revenue model or customer. The framework is MIT-licensed and reports are published openly, so it may behave more like a research institute than a venture-scale business.
  • –Co-founder change within about three months of incorporation. Aris Zhu now lists the role as previous on X, YC shows a single active founder, and no source opened explains why.
  • –Independence and revenue may conflict if payers are the labs and robot makers being scored. Reports on OpenAI and Anthropic models sit on subdomains named after those labs.

Could not confirm

  • ?The company says Inspect Robots has 97k+ downloads, not reproduced. Pypistats shows 49,756 for the inspect-robots package including mirrors, 7 July to 1 October 2026.
  • ?The company reports 6M+ research views and sign-ups from 200+ institutions, not independently confirmed.
  • ?When and why Aris Zhu left is not stated in any source opened.

SourcesRobocurve blog, seed announcement, 14 September 2026 ↗Y Combinator company page ↗Robocurve homepage ↗Aris Zhu LinkedIn post, 17 July 2026 ↗Ankit Gupta X article, 23 July 2026 ↗Aris Zhu X profile ↗GitHub, robocurve/inspect-robots ↗pypistats, inspect-robots downloads ↗SEC EDGAR full-text search, Robocurve ↗NBC News video, 23 September 2026 ↗

People

  • Aris ZhuSearch ↗

    Former co-founder (X bio reads "prev cofounder @robocurve"). YC partner Ankit Gupta's article lists Harvard's Computational Robotics Lab, Amazon Robotics, Amazon's AGI Lab and Yondu Robotics.

  • Edward SunSearch ↗

    Member of Technical Staff per the NBC News video description of 23 September 2026. No source opened states prior background.

  • Top contributor (381 commits)

    Founder and CEO of Robocurve. MA Statistics and BA CS/Math from Harvard. Jay builds evals of frontier AI and forecasts its progress. Previously Research Fellow at MATS, researcher at the UK AI Security Institute, and top contributor to Inspect Evals, the UK government's AI eval framework. Published at ACM EC, ICML, ACL, and EMNLP. Called the 2024 election correctly in all 50 states. Won a Rhodes Scholarship. Gold medal at the International Olympiad on Astronomy and Astrophysics.

  • Top contributor (71 commits)

  • Top contributor (41 commits)

    Bennett University | FullStack BU Team Member|

    Passionate Full Stack Developer | Tech Enthusiast | Problem Solver | Exploring scalable systems, cloud infrastructure, and modern web development.

Score

Momentum47

 

Code, attention, traffic

Thesis fit99

× 0.99

Matched on embodied ai, humanoid, industrial, manufacturing, physical ai

Earliness89

× 0.93

Already visible: web traffic, GitHub stars

Team1

× 0.85

5 named people, none from a top lab or company

47 × 0.99 × 0.93 × 0.85 = 37

Edge, 12 weeks