MSQA - Japanese: leaderboard
Metric: F-score (%; harmonic mean of CO, the share of fully correct answers, and CGA, the correct share of attempted answers, over the 83 Japanese-language 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 February 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 52.8 |
| 2 | GPT-5.5 | 48 |
| 3 | Claude Opus 4.7 | 42 |
| 4 | GPT-5.4 | 38.6 |
| 5 | Claude Opus 4.6 | 35.5 |
| 6 | DeepSeek V4 Pro | 34.9 |
| 7 | Seed 2.1 Pro | 31.8 |
| 8 | GPT-5.2 (High) | 30.1 |
| 9 | Seed 2.0 Pro (High) | 27.6 |
| 10 | Seed 2.0 Pro (Medium) | 25.9 |
| 11 | Gemini 2.5 Flash | 23.7 |
| 12 | Seed 2.0 Lite | 22.5 |
| 13 | DeepSeek V3.2 | 21.8 |
| 14 | Seed 2.1 Turbo | 21.8 |
| 15 | GLM-5 | 17.7 |
Interactive version: theaggregate.ai/benchmark?slug=msqa-japanese · How It Works · Data refreshed daily, snapshot 2026-09-29.