StatABench - Confidence Interval: leaderboard
StatABench (2026) tests LLM statistical analysis with tool use: Stat-Closed questions (multiple choice, fill in the blank, decision making and dataset-based practical application) are answered with a toolkit of statistical functions through LangChain MCP. This board is accuracy in percent on the 11 confidence interval questions; higher is better.
Metric: Accuracy (%) on the 11 confidence interval questions of Stat-Closed; Stat-Closed questions (multiple choice, fill in the blank, decision making, practical application with datasets) answered with the SAToolKit statistical functions through the LangChain MCP baseline tool-use setting, temperature 0; direct matching for closed answers and an LLM judge for open-form answers; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 | 63.6 | #716 |
| 2 | GPT-5.1 | 54.5 | #229 |
| 3 | Claude Sonnet 4.5 | 45.5 | #250 |
| 4 | Qwen 2.5 72B Instruct | 45.5 | #1092 |
| 5 | Qwen 2.5 32B Instruct | 27.3 | #1215 |
| 6 | Llama 3.1 8B Instruct | 18.2 | #4246 |
| 7 | Qwen 2.5 7B Instruct | 18.2 | #2564 |
| 8 | Qwen 3 8B (Non-reasoning) | 18.2 | #1830 (Qwen 3 8B) |
| 9 | GPT-4o Mini | 9.1 | #1623 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=statabench-confidence-interval · How It Works · Data refreshed daily, snapshot 2026-11-13.