Bagel Labs' Physical AI Model Ranks No. 3 Worldwide on the RoboDojo Benchmark
Reported by TMCnet and Tech Times2 sources ↓
TORONTO, Nov. 19, 2026 /PRNewswire/ -- Bagel Labs today announced Bagel-Robotics-1, an open-weight model ranked No. 3 worldwide among robot-control models on the RoboDojo benchmark, a global measure ...
Key points
- It completed 3.67% of open-task episodes, compared with 31.00% for GPT-6 Astra
- Bagel scored higher than Astra on partial scores in the four suites named in the company release
- The live leaderboard could not be loaded to independently confirm the ranking
Sources
Details
Tech Times reports that RoboDojo's news feed listed Bagel-Robotics-1 on Nov. 18 with an average score of 38.65 and a 31.91% success rate. The outlet could not load the live leaderboard to confirm the listing. The ranking is among standalone robot policies, not all systems displayed by the benchmark.
A comparison of benchmark results found Bagel completed 44 of 1,200 open-ended task episodes, a 3.67% success rate, while GPT-6 Astra recorded 31.00%. Bagel's results were better on partial scores in the four suites named in its release, but Astra had a 2.17-point advantage in success rate on Generalization; the release did not specify which metric its comparison used.
Bagel's own benchmark submission reports a 38.36 score and 31.59% success rate across three seeds, slightly below the leaderboard figures. Tech Times says the submission's pull request was still open on Nov. 20 and included later replacement layouts and episode changes; the sources reviewed did not explain the difference between the submission and leaderboard results.
Written from coverage as of