ActTraitBench: leaderboard
While Large Language Models (LLMs) can convincingly simulate personas in explicit self-reports, they often deviate in implicit behavioral decisions, revealing a substantial Knowledge-Decision Gap (G_KD). Existing benchmarks struggle to measure this asymmetry due to limited construct validity, multi-dimensional entanglement, and distributional biases in LLM-based evaluation. To address
Metric: G_KD (Global Knowledge-Decision Gap) (self-reported). Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 1.7B | 0.19 | #1196 |
| 2 | Qwen 3 8B | 0.23 | #674 |
| 3 | GPT-4o | 0.28 | #343 |
| 4 | DeepSeek V3 | 0.47 | #321 |
| 5 | Claude Sonnet 4.6 | 0.51 | #90 |
| 6 | DeepSeek V4 Pro | 0.57 | #102 |
| 7 | DeepSeek V3.2 | 0.59 | #205 |
| 8 | DeepSeek V4 Flash | 0.65 | #137 |
| 9 | DeepSeek V3.1 | 0.75 | #264 |
| 10 | Qwen 3 32B | 0.85 | #443 |
| 11 | Qwen 3 235B A22B (Thinking) | 1.54 | #313 (Qwen 3 235B A22B) |
| 12 | GLM-5 | 1.79 | #141 |
| 13 | Gemini 3.1 Pro (Preview) | 1.83 | #58 |
| 14 | MiniMax-M2.5 | 2.17 | #303 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=acttraitbench · How It Works · Data refreshed daily, snapshot 2026-10-20.