Vals AI Robo-bench (Robo-bench)
We show this table for reference; we do not rank on it.
Tests agents controlling a simulated robot arm and humanoid through motor commands, sensor readings, and camera images.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Mean task score on Robo-bench — October 13, 2026
We mirror the published mean task score view for Robo-bench. GPT-6.1 Sol leads the public snapshot at 28.33%, followed by GPT-6 Astra (23.33%) and Claude Opus 5.5 (14.58%). We do not use these results to rank models overall.
GPT-6.1 Sol
OpenAI
codex · max reasoning
codex
GPT-6 Astra
OpenAI
codex · max reasoning
codex
Claude Opus 5.5
Anthropic
claude-code · max reasoning
claude-code
7 modelsAgenticCurrentDisplay onlyUpdated October 13, 2026
Mean task score table (7 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-6.1 SolOpenAIcodex · max reasoningcodex | 28.33% | Not reported | Closed |
| 2 | GPT-6 AstraOpenAIcodex · max reasoningcodex | 23.33% | Not reported | Closed |
| 3 | Claude Opus 5.5Anthropicclaude-code · max reasoningclaude-code | 14.58% | Not reported | Closed |
| 4 | Muse Wonder FullMetamuse-code · max reasoningmuse-code | 10.42% | Not reported | Closed |
| 5 | Claude Fable 5.1Anthropicclaude-code · max reasoningclaude-code | 5.83% | Not reported | Closed |
| 6 | Gemini 4 ArgonGoogleantigravity · high reasoningantigravity | 0.83% | Not reported | Closed |
| 7 | Grok 4.7xAIgrok-build · xhigh reasoninggrok-build | 0.21% | Not reported | Closed |
How to read this leaderboard
Operator receipt: 7 sourced rows are currently displayable on this page; the leading published row is GPT-6.1 Sol at 28.33%.
Honest limit: These are model-and-agent results in MuJoCo simulations, not measurements on physical robots. The mean episode score is not the percentage of tasks fully solved. Most tasks are held out.
How Robo-bench is shown here
We mirror Vals AI’s Robo-bench results across 13 robot-arm tasks and 11 humanoid tasks in MuJoCo. Each episode lasts three minutes, and deliberation uses part of that time. Scores average task-level ratings from 0 to 100; they are not the share of tasks fully solved.
Each model runs in its provider-specific coding agent at the highest supported reasoning setting. These simulation results stay display only and do not enter model rankings or establish performance on physical robots.
Snapshot
The published Robo-bench snapshot places GPT-6.1 Sol first at 28.33%. The third row is 13.75 points behind. The broader top-10 range is 28.13 points, so the table still separates the published systems.
7 models have been evaluated on Robo-bench. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Robo-bench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Robo-bench
Year
2026
Tasks
24 simulated robot-control tasks
Format
Mean task score out of 100
Difficulty
Real-time perception, planning, and motor control
Published scores average 24 tasks scored from 0 to 100: 13 arm tasks and 11 humanoid tasks. Each episode runs for three minutes in real time, including model deliberation. Models use their provider-specific coding agents at the highest supported reasoning setting. Results are display only and excluded from model rankings.
Freshness and provenance
Version
Robo-bench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Robo-bench measure?
Tests agents controlling a simulated robot arm and humanoid through motor commands, sensor readings, and camera images.
Which model leads the published Robo-bench snapshot?
GPT-6.1 Sol currently leads the published Robo-bench snapshot with 28.33% mean task score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Robo-bench?
The October 13, 2026 snapshot contains 7 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 5,600+ readers.
One email each week. Unsubscribe anytime.