DTBench - Multi-hop Reasoning: leaderboard
Metric: Sub-capability-specific success rate (%) on DTBench: the share of ground-truth cells labelled multi-hop reasoning (the value chains several facts) that the model extracts exactly (after normalization and row alignment) when filling a target table schema from a long synthesized document; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 56.44 | #91 |
| 2 | GPT-5 Mini | 52.58 | #176 |
| 3 | DeepSeek V3.2 | 39.69 | #198 |
| 4 | Qwen 3 32B | 29.12 | #424 |
| 5 | Qwen 3 4B | 7.73 | #823 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=dtbench-multi-hop-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-11.