One Word, Different Action - Atomic Accuracy: leaderboard
One Word, Different Action (2026) grounds language instructions in 85 physical decision anchors recorded with a real robot, and the model must return the one executable decision an instruction implies. This board covers 3,952 atomic instances that each apply a single constraint: action inhibition, set exclusion, cardinality or temporal ordering. Exact-match accuracy in percent; higher is better.
Metric: Executable-decision accuracy (%; exact match of the executable decision (object, object set or ordered action sequence) against deterministic ground truth over 85 physical decision anchors collected with a real robot; 3,952 atomic instances in 988 task families, one constraint each (inhibition, exclusion, cardinality or temporal ordering); temperature 0, at most 512 new tokens, no chain of thought requested; an unmappable output counts as wrong). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5 FP8 | 99.4 | #260 |
| 2 | Mistral 7B Instruct (v0.3) | 97.8 | #6698 |
| 3 | OLMo-2-1124-7B-Instruct | 87.8 | #7377 |
| 4 | Qwen 2.5 7B Instruct | 84.4 | #2583 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=one-word-different-action-atomic-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-12.