MSQA - History and Collective Memory: leaderboard
Metric: F-score (%; harmonic mean of CO, the share of fully correct answers, and CGA, the correct share of attempted answers, over the 261 questions of this cultural dimension; free-form answers to natively sourced culture-specific short questions asked in their own language, five runs per model, scored by a Gemini-3.1-Pro gold-answer containment judge). Source: arxiv.org. Saturation forecast: Around May 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 68.5 |
| 2 | GPT-5.5 | 52.5 |
| 3 | Claude Opus 4.7 | 51 |
| 4 | DeepSeek V4 Pro | 50.8 |
| 5 | Claude Opus 4.6 | 48.7 |
| 6 | Seed 2.1 Pro | 47.4 |
| 7 | GPT-5.4 | 45.9 |
| 8 | Seed 2.0 Pro (High) | 45.1 |
| 9 | GLM-5 | 42.2 |
| 10 | GPT-5.2 (High) | 40.7 |
| 11 | Seed 2.1 Turbo | 39.9 |
| 12 | Seed 2.0 Pro (Medium) | 39.7 |
| 13 | Seed 2.0 Lite | 39.3 |
| 14 | Kimi K2.6 | 38.4 |
| 15 | Gemini 2.5 Flash | 38.2 |
Interactive version: theaggregate.ai/benchmark?slug=msqa-history-and-collective-memory · How It Works · Data refreshed daily, snapshot 2026-09-29.