DEPLOYDatabase

Company

Robocurve

Robocurve is a San Francisco Public Benefit Corporation that runs independent real-world benchmarks of how well AI models and robots perform physical tasks.

Founded
2026
Status
Private

Overview

Robocurve is a San Francisco Public Benefit Corporation that runs independent real-world benchmarks of how well AI models and robots perform physical tasks. It also builds Inspect Robots, an open-source evaluation framework.

Reality check

Cost per robot subtask (company-claimed, Oct 2026)

Estimated API cost for frontier LLMs to complete a RoboDojo-RC Tier 1 subtask fell from $2.36 in July to $0.38 in September 2026, halving about every 3 weeks.

Verified record

Verified deployments
None on file
Active incidents
None on file

Key facts

RoboDojo-RC Tier 1 benchmark (company-run)

Six real-world tasks on bimanual YAM arms, such as stacking bowls, capping a pen and sorting objects, scored on a progress rubric across 1,431 runs.

RoboDojo-RC Tier 1 top score (company-run, Oct 2026)

Gemini 4 Argon (high) leads with a 48.2% mean score over 120 attempts, ahead of GPT-6.1 Sol (high) at 41.7%.

RoboHarm v1 benchmark (company-run)

Tests whether robot policies refuse unsafe instructions; models refused for safety in 26.4% of 698 runs.

Inspect Robots framework (company-stated)

Open-source, MIT-licensed tool to evaluate VLAs and LLMs on real robots, with simulation support and full trace logging.

Y Combinator batch (company-stated)

Summer 2026

Data & sources

Web sources

9

9 sources backing this record.View all →

Current leadership (1)

Founders (2)

See all founders across the registry →

Safety record

No incidents on record for Robocurve.

Only active incidents are counted. Retracted incidents are excluded from this summary but remain reachable at their canonical URLs.

Full safety record: incidents, sourcing, and exposure data →

Mentioned on (2)

Registry pages whose text names Robocurve.

People (2)

Aris Zhu, Jay Chooi