Benchmark Almanac

RoboDojo-RC Tier 1

Dream-1 59.3% tops its leaderboard Gemini 4 Argon 48.2% leads the models we track
Not in model cards

Six real RoboDojo tasks on I2RT YAM arms, 20 attempts each. Robot policies and tool-calling LLMs scored by rubric progress.

One team runs every attempt on its own rigs and publishes per-trial score, cost, time, logs and video. Best entry Dream-1 59.3% progress (trained on YAM). Robocurve says best LLM score rose 15% to 43% in two months.

Made by
Robocurve PBC
Released
September 2026
Kind
Real robots
Level
Everyday
Scored
Graded by people
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
September 2026
Scores refreshed
November 12th, 2026
11
Entries on its own leaderboard, 9 models we track

Dream-1 tops the leaderboard with 59.3%

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Gemini 4 Argon, at 48.2%.

  1. 59.3 Dream-1
  2. 54.5 Gemini 4 Carbon (early access)
  3. 48.2 Gemini 4 Argon
  4. 42.5 GPT-6 Astra
  5. 41.7 GPT-6.1 Sol
  6. 38.2 Claude Opus 5.5
  7. 36.2 Gemini 3.8 Flash
  8. 22.0 GPT-6 Sol
  9. 18.9 MolmoAct 2
  10. 14.9 GPT-5.6 Sol
  11. 9.51 GR00T N1.7
  1. 1 Dream-1 hosted 59.3%
  2. 2 Gemini 4 Carbon (early access) high; standard 54.5%
  3. 3 Gemini 4 Argon high; priority 48.2%
  4. 4 GPT-6 Astra high; fast 42.5%
  5. 5 GPT-6.1 Sol high; fast 41.7%
  6. 6 Claude Opus 5.5 high; fast 38.2%
  7. 7 Gemini 3.8 Flash high; standard 36.2%
  8. 8 GPT-6 Sol high; fast 22%
  9. 9 MolmoAct 2 self-hosted 18.9%
  10. 10 GPT-5.6 Sol high; fast 14.9%
  11. 11 GR00T N1.7 self-hosted 9.51%

Source: Official RoboDojo-RC Tier 1 leaderboard

#EntryScoreSettingDateSource
1 Dream-1 59.3% hosted Official leaderboard
2 Gemini 4 Carbon (early access) 54.5% high; standard Official leaderboard
3 Gemini 4 Argon 48.2% high; priority Sep 2026 Official leaderboard
4 GPT-6 Astra 42.5% high; fast Sep 2026 Official leaderboard
5 GPT-6.1 Sol 41.7% high; fast Sep 2026 Official leaderboard
6 Claude Opus 5.5 38.2% high; fast Sep 2026 Official leaderboard
7 Gemini 3.8 Flash 36.2% high; standard Sep 2026 Official leaderboard
8 GPT-6 Sol 22% high; fast Sep 2026 Official leaderboard
9 MolmoAct 2 18.9% self-hosted May 2026 Official leaderboard
10 GPT-5.6 Sol 14.9% high; fast Jul 2026 Official leaderboard
11 GR00T N1.7 9.51% self-hosted Apr 2026 Official leaderboard

Robotics Capabilities vs Time

Mean score against wall-clock time per task attempt, 120 attempts per configuration; robot policies (VLAs) always run to their 3,600-step cap. The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 20% 40% 60% 101001,00010K Time per task attempt (s), log scale, lower is better Mean score GR00T N1.7 (self-hosted): Mean score 9.51%, time per task attempt 56 s GPT-5.6 Sol (high; fast): Mean score 14.9%, time per task attempt 218 s Gemini 3.8 Flash (high; standard): Mean score 36.2%, time per task attempt 433 s GPT-6 Astra (high; fast): Mean score 42.5%, time per task attempt 238 s GPT-6 Astra (medium; standard): Mean score 36.7%, time per task attempt 251 s Claude Opus 5.5 (high; fast): Mean score 38.2%, time per task attempt 186 s Claude Opus 5.5 (medium; standard): Mean score 36%, time per task attempt 269 s GPT-6 Sol (high; fast): Mean score 22%, time per task attempt 224 s GPT-6.1 Sol (high; fast): Mean score 41.7%, time per task attempt 242 s Gemini 4 Argon (high; priority): Mean score 48.2%, time per task attempt 1,230 s Gemini 4 Carbon (early access) (high; standard): Mean score 54.5%, time per task attempt 1,923 s. Not in our model list MolmoAct 2 (self-hosted): Mean score 18.9%, time per task attempt 48 s Dream-1 (hosted): Mean score 59.3%, time per task attempt 92 s. Not in our model list Dream-1 (hosted)MolmoAct 2 (self-hosted) 0% 20% 40% 60% 101001,00010K Time per task attempt (s), log scale, lower is better Mean score GR00T N1.7 (self-hosted): Mean score 9.51%, time per task attempt 56 s GPT-5.6 Sol (high; fast): Mean score 14.9%, time per task attempt 218 s Gemini 3.8 Flash (high; standard): Mean score 36.2%, time per task attempt 433 s GPT-6 Astra (high; fast): Mean score 42.5%, time per task attempt 238 s GPT-6 Astra (medium; standard): Mean score 36.7%, time per task attempt 251 s Claude Opus 5.5 (high; fast): Mean score 38.2%, time per task attempt 186 s Claude Opus 5.5 (medium; standard): Mean score 36%, time per task attempt 269 s GPT-6 Sol (high; fast): Mean score 22%, time per task attempt 224 s GPT-6.1 Sol (high; fast): Mean score 41.7%, time per task attempt 242 s Gemini 4 Argon (high; priority): Mean score 48.2%, time per task attempt 1,230 s Gemini 4 Carbon (early access) (high; standard): Mean score 54.5%, time per task attempt 1,923 s. Not in our model list MolmoAct 2 (self-hosted): Mean score 18.9%, time per task attempt 48 s Dream-1 (hosted): Mean score 59.3%, time per task attempt 92 s. Not in our model list Dream-1 (hosted)MolmoAct 2 (self-hosted)
ModelSettingMean scoreTime per task attempt
Dream-1 hosted 59.3% 92 s
Gemini 4 Carbon (early access) high; standard 54.5% 1,923 s
Gemini 4 Argon high; priority 48.2% 1,230 s
GPT-6 Astra high; fast 42.5% 238 s
GPT-6.1 Sol high; fast 41.7% 242 s
Claude Opus 5.5 high; fast 38.2% 186 s
GPT-6 Astra medium; standard 36.7% 251 s
Gemini 3.8 Flash high; standard 36.2% 433 s
Claude Opus 5.5 medium; standard 36% 269 s
GPT-6 Sol high; fast 22% 224 s
MolmoAct 2 self-hosted 18.9% 48 s
GPT-5.6 Sol high; fast 14.9% 218 s
GR00T N1.7 self-hosted 9.5% 56 s

The best score, 48.2%, costs $2.34 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

NVIDIA Ai2 Google DeepMind OpenAI Cost frontier One model at each effort level
0% 10% 20% 30% 40% 50% $0.01 $0.10 $1.00 $10 Cost per task, log scale Gemini 4 Argon, high; priority GPT-6.1 Sol, high; fast GPT-6 Astra, medium; standard MolmoAct 2, self-hosted GR00T N1.7, self-hosted 0% 10% 20% 30% 40% 50% $0.01 $0.10 $1.00 $10 Cost per task, log scale Gemini 4 Argon, high; priority GPT-6.1 Sol, high; fast

Source: Official RoboDojo-RC Tier 1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon high; priority 48.2% $2.34
GPT-6 Astra high; fast 42.5% $3.48
GPT-6.1 Sol high; fast 41.7% $2.15 9,143
Claude Opus 5.5 high; fast 38.2% $2.23
GPT-6 Astra medium; standard 36.7% $1.14
Gemini 3.8 Flash high; standard 36.2% $0.66
Claude Opus 5.5 medium; standard 36% $0.90
GPT-6 Sol high; fast 22% $0.95
MolmoAct 2 self-hosted 18.9% $0.07
GPT-5.6 Sol high; fast 14.9% $1.77
GR00T N1.7 self-hosted 9.51% $0.05

4 models have held number one on the official leaderboard since its release in September 2026

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

Ai2 Google DeepMind OpenAI
Own leaderboard 4 leaders
Gemini 4 Argon 48.2%

Source: Official RoboDojo-RC Tier 1 leaderboard

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% 2026 Near the ceiling (85%) Gemini 4 Argon GPT-6 Astra Gemini 3.8 Flash MolmoAct 2 GR00T N1.7 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Gemini 3.8 Flash

Source: Official RoboDojo-RC Tier 1 leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Robotics

All 103 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%
  • RoboDojo In use Generalist manipulation policies run 42 Isaac Sim tasks and 18 remote real-robot tasks. Scored by task score and success rate. GPT-6 Astra 29%
  • RoboLab In use Zero-shot DROID Franka policies in Isaac Lab on 120 tabletop tasks at three instruction levels, 10 episodes each. Success rate. FLUX 3 Action 42.9%
  • 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%
  • 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%