SCQA (No Search) - Multi-Hop High-Visibility: leaderboard
Metric: Set F1 (0-1; 20 multi-hop list questions on publicly documented facts; direct prompting without web search, temperature 0, ChatGPT 5.5 judge). Source: arxiv.org. Saturation forecast: Around April 2027. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 0.55 |
| 2 | DeepSeek V3.2 | 0.48 |
| 3 | Qwen 3 235B A22B | 0.44 |
| 4 | GLM-5 | 0.4 |
Interactive version: theaggregate.ai/benchmark?slug=scqa-no-search-multi-hop-high-visibility · How It Works · Data refreshed daily, snapshot 2026-09-25.