Benchmark Almanac

DataCenterBench Tier 1

GPT-6 Astra 40% leads since Oct 2026 Not in model cards

Frontier models watch through cameras and drive two real robot arms on tabletop tasks rebuilt from data-center equipment manuals. LLM judges score each try from photos.

Published 2026-10-28 with ten tasks, three tries each. Runs on Robocurve's inspect-robots harness (723 GitHub stars). 9 models in the Fireworks index.

Made by Robocurve Released Oct 2026 Real robots Tasks kept private Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
robocurve/inspect-robots 723 stars Leaderboard robocurve.org
9
Models scored on its own leaderboard

GPT-6 Astra leads with 40%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 40.0 GPT-6 Astra
  2. 38.3 GPT-6.1 Sol
  3. 35.0 Gemini 4 Argon
  4. 21.7 Claude Opus 5.5
  5. 16.7 Muse Wonder
  6. 15.0 Kimi K3
  7. 15.0 GLM-5.3-Flash
  8. 13.3 Grok 4.7
  9. 11.7 DeepSeek V4.1 Flash
  1. 1 GPT-6 Astra max, ultrafast 40%
  2. 2 GPT-6.1 Sol max, ultrafast 38.3%
  3. 3 Gemini 4 Argon high 35%
  4. 4 Claude Opus 5.5 max 21.7%
  5. 5 Muse Wonder high 16.7%
  6. 6 Kimi K3 high 15%
  7. 7 GLM-5.3-Flash high 15%
  8. 8 Grok 4.7 xhigh 13.3%
  9. 9 DeepSeek V4.1 Flash high 11.7%
#ModelScoreSettingDateSource
1 GPT-6 Astra 40% max, ultrafast Sep 2026 Official leaderboard
2 GPT-6.1 Sol 38.3% max, ultrafast Sep 2026 Official leaderboard
3 Gemini 4 Argon 35% high Sep 2026 Official leaderboard
4 Claude Opus 5.5 21.7% max Sep 2026 Official leaderboard
5 Muse Wonder 16.7% high Oct 2026 Official leaderboard
6 Kimi K3 15% high Jul 2026 Official leaderboard
7 GLM-5.3-Flash 15% high Aug 2026 Official leaderboard
8 Grok 4.7 13.3% xhigh Sep 2026 Official leaderboard
9 DeepSeek V4.1 Flash 11.7% high Sep 2026 Official leaderboard

The best score, 40%, costs $19 a task

Runs on the cost frontier score higher than any run that's cheaper. Gemini 4 Argon at high comes within 5 points of the best for $1.83 a task, 10 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Z.ai OpenAI Google DeepMind Meta Cost frontier One model at each effort level
0% 10% 20% 30% 40% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra, max, ultrafast GPT-6.1 Sol, max, ultrafast Gemini 4 Argon, high Muse Wonder, high GLM-5.3-Flash, high 0% 10% 20% 30% 40% $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6 Astra, max, ultrafast Gemini 4 Argon, high
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6 Astra max, ultrafast 40% $19 22,860
GPT-6.1 Sol max, ultrafast 38.3% $10 21,904
Gemini 4 Argon high 35% $1.83 102,960
Claude Opus 5.5 max 21.7% $18 230,136
Muse Wonder high 16.7% $1.52 29,011
Kimi K3 high 15% $0.60 34,970
GLM-5.3-Flash high 15% $0.01 49,118
Grok 4.7 xhigh 13.3% $0.60 16,737
DeepSeek V4.1 Flash high 11.7% $0.14 112,484

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard 1 leader
GPT-6 Astra 40%

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% Near the ceiling (85%) GPT-6 Astra Kimi K3 0% 20% 40% 60% 80% 100% Near the ceiling (85%) GPT-6 Astra Kimi K3

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Robotics

All 86 in Robotics
  • ERQA In use 400 multiple-choice visual questions from robot and egocentric footage on spatial, trajectory, action and state reasoning. Accuracy. Gemini Robotics ER 2 78.5%
  • RoboLab In use Zero-shot DROID Franka policies in Isaac Lab on 120 tabletop tasks at three instruction levels, 10 episodes each. Success rate. GR00T N1.7 8.58%
  • Human proximity (Google DeepMind robotics) In use Models judge from robot camera views whether a human is within 1 meter, a spatial awareness check for safe robot behavior. Scored by accuracy. Gemini Robotics ER 2 93%
  • Moment finding (Google DeepMind) In use Models find the moment in a robot video when an event happens. Scored by accuracy and mean absolute time distance in seconds. Gemini Robotics ER 2 91.3%
  • Progress classification (Google DeepMind) In use Models assign each frame of a robot video to one of five task-progress bands from 0 to 100 percent. Scored by accuracy. Gemini Robotics ER 2 57.4%
  • LIBERO Fading Simulated Franka tabletop tasks in four 10-task suites. Policies fine-tuned on 50 demos per task. Success rate over ~50 trials. MolmoAct 2 97.2%