EcoG-Bench (Frames + ASR): leaderboard
Metric: Eco-accuracy (%): share of referents whose option (what), 2D point on the last frame (where) and timestamp inside the gesture-stroke window (when) are all correct, over EcoG-Bench's 811 egocentric clips (English and Chinese) with underspecified deictic instructions resolved by co-speech pointing, levels L1-L4 with one to four referents, temperature 0 and JSON output (unparsable answers score as wrong); the model sees frames sampled at 2 fps with millisecond timestamps plus a verified ASR transcript without word timings; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 3.3 | #176 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=ecog-bench-frames-plus-asr · How It Works · Data refreshed daily, snapshot 2026-10-11.