SCQA (No Search) - Multi-Hop Low-Visibility: leaderboard
Metric: Graph F1 (0-1; 0.6 x entity F1 + 0.4 x relation F1 against a reference supply-chain graph; 20 multi-hop questions on facts found only in niche sources; direct prompting without web search, temperature 0, ChatGPT 5.5 judge). Source: arxiv.org. Saturation forecast: Around March 2027. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 0.63 |
| 2 | GLM-5 | 0.33 |
| 3 | DeepSeek V3.2 | 0.27 |
| 4 | Qwen 3 235B A22B | 0.25 |
Interactive version: theaggregate.ai/benchmark?slug=scqa-no-search-multi-hop-low-visibility · How It Works · Data refreshed daily, snapshot 2026-09-25.