DataCenterBench Tier 1

DataCenterBench Tier 1 is a robotics benchmark of ten tabletop tasks rebuilt from steps in data-center equipment manuals, such as operating controls and connecting cables, done by models driving two robot arms. Each model tries every task three times, LLM judges score the tries from photos, and the score is the mean of the three runs’ scores, with Robocurve’s standard error across runs.

Last updated · Source leaderboard checked daily

Main Insights

Results

DataCenterBench Tier 1 results, ranked by score (avg@3), highest first. Last updated .
RankModelProviderScore (avg@3)Cost (USD / task)Duration (seconds / task)
1GPT-6 AstraOpenAI40.0%$3.19215
2GPT-6.1 SolOpenAI38.3%$1.75183
3Gemini 4 ArgonGoogle35.0%$1.831,877
4Claude Opus 5.5Anthropic21.7%$8.83892
5Muse Wonder 1.0Meta16.7%$1.52509
6GLM 5.3 FlashZ.ai15.0%$0.086976
7Kimi K3Moonshot AI15.0%$1.171,230
8Grok 4.7xAI13.3%$0.60305
9DeepSeek V4.1 FlashDeepSeek11.7%$0.24425