What is the top score on RoboDojo (LLM as robot policy)?

As of 13 Nov 2026, the highest score on RoboDojo (LLM as robot policy) is 29.0 by GPT-6 Astra (result dated 3 Nov 2026), ahead of GPT-5.5 at 1.1. Scores come from RoboDojo and are re-read daily.

Which AI model leads RoboDojo (LLM as robot policy)?

GPT-6 Astra leads RoboDojo (LLM as robot policy) with 29.0 as of 13 Nov 2026 (result dated 3 Nov 2026), ahead of GPT-5.5 at 1.1.

How many models have RoboDojo (LLM as robot policy) results?

2 model configurations have a RoboDojo (LLM as robot policy) result on BenchLeader, all taken from RoboDojo.

Who runs RoboDojo (LLM as robot policy) and how often is it updated?

RoboDojo (LLM as robot policy) is published by RoboDojo. BenchLeader re-reads the published results once a day and records the date each result was published.

Does RoboDojo (LLM as robot policy) count toward the BenchLeader Index?

No. RoboDojo (LLM as robot policy) is shown for reference but left out of the composite index.

RoboDojo (LLM as robot policy) leaderboard

As of 13 Nov 2026, the top score on RoboDojo (LLM as robot policy) is 29.0 by GPT-6 Astra (OpenAI), ahead of GPT-5.5 at 1.1, across 2 model configurations with a published result (RoboDojo, result dated 3 Nov 2026). RoboDojo tests robot manipulation in simulation: 42 tasks such as stacking, sweeping and opening, grouped into generalization, precision, long-horizon, memory and open-ended skills. Here a general LLM acts as the robot's policy, choosing the actions itself with no robot-specific training (the RoboDojo team’s “LLM as policy” setting). RoboDojo (LLM as robot policy) is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 13 Nov 2026)

GPT-6 Astra holds the top RoboDojo (LLM as robot policy) score at 29.0 (RoboDojo, 3 Nov 2026).
GPT-5.5 is second on RoboDojo (LLM as robot policy) at 1.1 (RoboDojo, 3 Nov 2026).
RoboDojo (LLM as robot policy): A general LLM controls a simulated robot arm directly, with no robot training, across 42 manipulation tasks (generalization, precision, long-horizon, memory, open-ended). The score averages per-task scores from 0 to 100; the success rate is in each row's details (RoboDojo).
2 model configurations have a published RoboDojo (LLM as robot policy) result on BenchLeader (BenchLeader, 13 Nov 2026).
RoboDojo (LLM as robot policy) does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

A general LLM controls a simulated robot arm directly, with no robot training, across 42 manipulation tasks (generalization, precision, long-horizon, memory, open-ended). The score averages per-task scores from 0 to 100; the success rate is in each row's details.

Published by: RoboDojo
Category: Agents & tools
Index weight: Reference only
Models: 2
Data as of: 13 Nov 2026

MIT — robodojo-benchmark/RoboDojo (https://robodojo-benchmark.com). LLM results from the RoboDojo team (arXiv 2609.24170). Not endorsed by RoboDojo.

What the test looks like

RoboDojo tests robot manipulation in simulation: 42 tasks such as stacking, sweeping and opening, grouped into generalization, precision, long-horizon, memory and open-ended skills. Here a general LLM acts as the robot's policy, choosing the actions itself with no robot-specific training (the RoboDojo team’s “LLM as policy” setting).

How it is scored

Each task gives a 0 to 100 score and a success rate over 50 episodes; the board averages both across tasks. Shown is the average score, with the success rate and rank among every policy on the board (mostly robot-trained models) in the details.

What to keep in mind

Only a few LLMs have been run, all by the RoboDojo team, and the harness and prompting are theirs. The robot-trained policies it is compared against are not language models, so they do not appear here. Kept out of the index.

Top 15

1GPT-6 Astra29.0
2GPT-5.51.1

2 of 2

#
1	GPT-6 AstraOpenAI	29.0	66.8	2026-11-03
2	GPT-5.5OpenAI	1.1	61.2	2026-11-03

AI agent benchmark leaderboard · All LLM benchmarks

Top head-to-heads on RoboDojo (LLM as robot policy)

GPT-6 Astra vs GPT-5.5

Cite as: BenchLeader, “RoboDojo (LLM as robot policy) leaderboard”, data as of 13 Nov 2026, https://www.benchleader.com/benchmarks/robodojo_sim.

RoboDojo (LLM as robot policy): questions

What does RoboDojo (LLM as robot policy) measure?: RoboDojo tests robot manipulation in simulation: 42 tasks such as stacking, sweeping and opening, grouped into generalization, precision, long-horizon, memory and open-ended skills. Here a general LLM acts as the robot's policy, choosing the actions itself with no robot-specific training (the RoboDojo team’s “LLM as policy” setting). Scores are reported in index; higher is better.
What is the top score on RoboDojo (LLM as robot policy)?: As of 13 Nov 2026, the highest score on RoboDojo (LLM as robot policy) is 29.0 by GPT-6 Astra (result dated 3 Nov 2026), ahead of GPT-5.5 at 1.1. Scores come from RoboDojo and are re-read daily.
Which AI model leads RoboDojo (LLM as robot policy)?: GPT-6 Astra leads RoboDojo (LLM as robot policy) with 29.0 as of 13 Nov 2026 (result dated 3 Nov 2026), ahead of GPT-5.5 at 1.1.
How many models have RoboDojo (LLM as robot policy) results?: 2 model configurations have a RoboDojo (LLM as robot policy) result on BenchLeader, all taken from RoboDojo.
Who runs RoboDojo (LLM as robot policy) and how often is it updated?: RoboDojo (LLM as robot policy) is published by RoboDojo. BenchLeader re-reads the published results once a day and records the date each result was published.
Does RoboDojo (LLM as robot policy) count toward the BenchLeader Index?: No. RoboDojo (LLM as robot policy) is shown for reference but left out of the composite index.