IntentGrasp - Gem Set: leaderboard
The Gem Set of IntentGrasp (2026), 470 intent-understanding items from 12 domains that every open-weight model failed in its first run, balanced across domains. Each item asks for the intended meaning among options; the score is instance F1 in percent, averaged over four option orders; higher is better.
Metric: Instance F1 (%) on the 470-item Gem Set, mean of four option-order runs (no shuffle and shuffle seeds 7, 42 and 365), zero-shot, API default settings; higher is better. Source: arxiv.org. Saturation forecast: Around December 2027. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 24.67 | #152 |
| 2 | Gemini 3.1 Pro (Preview) | 21.51 | #99 |
| 3 | Claude Opus 4.7 | 16.55 | #85 |
| 4 | Gemini 3.1 Flash Lite (Preview) | 11.77 | #417 |
| 5 | GPT-5.4 | 11.71 | #135 |
| 6 | Claude Sonnet 4.6 | 10.91 | #173 |
| 7 | GPT-5.4 Nano | 8.68 | #651 |
| 8 | GPT-5.4 Mini | 7.08 | #406 |
| 9 | Claude Haiku 4.5 | 5.76 | #583 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=intentgrasp-gem-set · How It Works · Data refreshed daily, snapshot 2026-11-21.