MSQA: leaderboard
Metric: F-score (%; harmonic mean of CO, the share of fully correct answers, and CGA, the correct share of attempted answers, over all 1,064 questions; 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 April 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 68.7 |
| 2 | GPT-5.5 | 55.6 |
| 3 | Claude Opus 4.6 | 52.8 |
| 4 | GPT-5.4 | 50.9 |
| 5 | DeepSeek V4 Pro | 50.8 |
| 6 | Seed 2.1 Pro | 50.4 |
| 7 | Claude Opus 4.7 | 49 |
| 8 | GPT-5.2 (High) | 45.4 |
| 9 | Seed 2.0 Pro (High) | 44.3 |
| 10 | GLM-5 | 42.1 |
| 11 | Seed 2.0 Pro (Medium) | 41.2 |
| 12 | Seed 2.1 Turbo | 40.8 |
| 13 | Gemini 2.5 Flash | 40.6 |
| 14 | DeepSeek V3.2 | 40.4 |
| 15 | Qwen 3.5 Plus (Thinking) | 40.4 |
Interactive version: theaggregate.ai/benchmark?slug=msqa · How It Works · Data refreshed daily, snapshot 2026-09-29.