SABench - Goal Satisfaction: leaderboard
SABench measures satisfaction of deterministic goals in stateful tool tasks. Its 201-task set includes 45 ambiguous tasks with no stated goals, which receive goal credit automatically. Deterministic goal-satisfaction rate on 201 tasks (%).
Metric: Deterministic goal-satisfaction rate on201 tasks (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 95.5 | #103 |
| 2 | Claude Sonnet 4.6 | 92 | #96 |
| 3 | Qwen 3.5 Plus | 90.5 | #131 |
| 4 | Kimi K2.5 | 90 | #148 |
| 5 | GPT-5.4 Mini | 89.5 | #213 |
| 6 | GLM-5 | 89 | #145 |
| 7 | GPT-5.4 | 88.1 | #83 |
| 8 | Claude Opus 4.6 | 87.6 | #65 |
| 9 | DeepSeek V3.2 | 87.6 | #205 |
| 10 | Gemini 3.1 Pro (Preview) | 85.6 | #61 |
| 11 | MiniMax-M2.5 | 85.6 | #303 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sabench-goal-satisfaction · How It Works · Data refreshed daily, snapshot 2026-10-29.