QuantSightBench: leaderboard
Numerical forecasting benchmark evaluating whether LLMs produce calibrated 90% prediction intervals for 1,000 real-world questions under zero-shot, grounded, and agentic retrieval settings.
Metric: Coverage (%). Source: quantsightbench.com. Source: benchmarklist.com. Saturation forecast: Around January 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 79.1 | #58 |
| 2 | Grok 4 | 76.38 | #181 |
| 3 | GPT-5.4 | 75.33 | #79 |
| 4 | GPT-5.1 | 74.59 | #135 |
| 5 | Claude Opus 4.6 | 73.6 | #63 |
| 6 | Claude Opus 4.5 | 71.66 | #83 |
| 7 | Claude Sonnet 4.5 | 67.96 | #142 |
| 8 | Kimi K2 (Thinking) | 65.79 | #243 (Kimi K2) |
| 9 | Gemini 3 Pro | 65.43 | #80 |
| 10 | GLM-4.7 | 62.69 | #192 |
| 11 | DeepSeek V3.2 | 61.48 | #205 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=quantsightbench · How It Works · Data refreshed daily, snapshot 2026-10-20.