Benchmark Almanac

EgoHome

Perceptron Egocentric 0.80 tops its leaderboard GPT-6 Astra 0.80 leads the models we track
Not in model cards

Models annotate 137 egocentric clips of household chores, about two hours in all, with subtasks, hand actions, hand captions and hand boxes. Segments are as short as 0.1 seconds. A model judge matches segment labels, and boxes are scored by centroid hits.

Perceptron released it with Perceptron Egocentric 1.0. Each annotation type gets its own score from 0 to 1, and the headline here is the subtask score, which the card plots against cost. Perceptron averaged its own model over several runs, and every other system ran once. The data is CC BY-NC 4.0, with a grant that lets anyone evaluate models on it and publish the results.

Made by
Perceptron
Released
November 2026
Kind
Video understanding
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
September 2026
Scores refreshed
November 14th, 2026
6
Entries on its own leaderboard, 3 models we track

Perceptron Egocentric tops the leaderboard with 0.80

Best score per entry from the benchmark's own leaderboard, read on November 14th, 2026. Where a model ran with several settings or harnesses, we keep its best. 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 GPT-6 Astra, at 0.80.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 0.80 Perceptron Egocentric
  2. 0.80 GPT-6 Astra
  3. 0.75 FLORA (NVIDIA)
  4. 0.69 Glimpse (Elorian)
  5. 0.66 Gemini 3.8 Flash
  6. 0.32 GPT-6.1 Sol
  1. 1 Perceptron Egocentric 8 fps, option 0.80
  2. 2 GPT-6 Astra 4 fps frames 0.80
  3. 3 FLORA (NVIDIA) 0.75
  4. 4 Glimpse (Elorian) 4 fps video 0.69
  5. 5 Gemini 3.8 Flash 4 fps video 0.66
  6. 6 GPT-6.1 Sol Midcentury's settings 0.32

Source: Official EgoHome leaderboard

#EntryScoreSettingDateSource
1 Perceptron Egocentric 0.80 8 fps, option Nov 2026 Official leaderboard
2 GPT-6 Astra 0.80 4 fps frames Sep 2026 Official leaderboard
3 FLORA (NVIDIA) 0.75 Nov 2026 Official leaderboard
4 Glimpse (Elorian) 0.69 4 fps video Nov 2026 Official leaderboard
5 Gemini 3.8 Flash 0.66 4 fps video Sep 2026 Official leaderboard
6 GPT-6.1 Sol 0.32 Midcentury's settings, Argus (Pantheon) Sep 2026 Official leaderboard

Subtask score against subtask cost per video-hour

The card plots each system’s subtask score against what one hour of video costs to annotate. 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 0.2 0.4 0.6 0.8 $1$10$100 Subtask cost per video-hour ($), log scale, lower is better Subtask score GPT-6 Astra (4 fps frames): Subtask score 0.8, subtask cost per video-hour $76 GPT-6 Astra (1 fps frames): Subtask score 0.78, subtask cost per video-hour $20 GPT-6 Astra: Subtask score 0.37, subtask cost per video-hour $24.2 GPT-6.1 Sol (Midcentury's settings): Subtask score 0.32, subtask cost per video-hour $7.76 FLORA (NVIDIA): Subtask score 0.75, subtask cost per video-hour $35.4. Not in our model list Gemini 3.8 Flash (4 fps video): Subtask score 0.66, subtask cost per video-hour $1.3 Perceptron Egocentric (8 fps, option): Subtask score 0.8, subtask cost per video-hour $15. Not in our model list Perceptron Egocentric (4 fps, default): Subtask score 0.79, subtask cost per video-hour $9. Not in our model list Glimpse (Elorian) (4 fps video): Subtask score 0.69, subtask cost per video-hour $2.8. Not in our model list Perceptron Egocentric (8 fps, option)Perceptron Egocentric (4 fps, default)Glimpse (Elorian) (4 fps video)Gemini 3.8 Flash (4 fps video) 0 0.2 0.4 0.6 0.8 $1$10$100 Subtask cost per video-hour ($), log scale, lower is better Subtask score GPT-6 Astra (4 fps frames): Subtask score 0.8, subtask cost per video-hour $76 GPT-6 Astra (1 fps frames): Subtask score 0.78, subtask cost per video-hour $20 GPT-6 Astra: Subtask score 0.37, subtask cost per video-hour $24.2 GPT-6.1 Sol (Midcentury's settings): Subtask score 0.32, subtask cost per video-hour $7.76 FLORA (NVIDIA): Subtask score 0.75, subtask cost per video-hour $35.4. Not in our model list Gemini 3.8 Flash (4 fps video): Subtask score 0.66, subtask cost per video-hour $1.3 Perceptron Egocentric (8 fps, option): Subtask score 0.8, subtask cost per video-hour $15. Not in our model list Perceptron Egocentric (4 fps, default): Subtask score 0.79, subtask cost per video-hour $9. Not in our model list Glimpse (Elorian) (4 fps video): Subtask score 0.69, subtask cost per video-hour $2.8. Not in our model list Perceptron Egocentric (8 fps, option)Perceptron Egocentric (4 fps, default)Glimpse (Elorian) (4 fps video)Gemini 3.8 Flash (4 fps video)
ModelSettingSubtask scoreSubtask cost per video-hour
Perceptron Egocentric 8 fps, option 0.8 $15
GPT-6 Astra 4 fps frames 0.8 $76
Perceptron Egocentric 4 fps, default 0.79 $9
GPT-6 Astra 1 fps frames 0.78 $20
FLORA (NVIDIA) 0.75 $35.39
Glimpse (Elorian) 4 fps video 0.69 $2.8
Gemini 3.8 Flash 4 fps video 0.66 $1.3
GPT-6 Astra Argus (Pantheon) 0.37 $24.17
GPT-6.1 Sol Midcentury's settings, Argus (Pantheon) 0.32 $7.76

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 0.80

Source: Official EgoHome 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
0.40 0.50 0.60 0.70 0.80 0.40 0.50 0.60 0.70 0.80

Source: Official EgoHome 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 Audio and video

All 233 in Audio and video
  • Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
  • LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization. Gemini 4 Argon 91.7%
  • MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
  • MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
  • FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better. Phi-4-multimodal 4.00%
  • VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%