NOVA-63
General QA
NOVA-63 (Native Omni-lingual Versatile Assessments of 63 disciplines) is a multiple-choice benchmark of natively written questions in 14 languages across 63 academic disciplines, built to measure cross-lingual consistency of knowledge. Scores report accuracy; the paper describes two edition sizes, so no fixed item count is asserted.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.5 397B A17B | 59.1 |
| Opus 4.6 | 59.1 |
| Qwen3.7-Max (source unspecified) | 59.0 |
| Qwen3.7-Plus (source unspecified) | 58.8 |
| Qwen3.5 122B A10B | 58.6 |
| Qwen3.5-27B | 58.1 |
| Qwen3.6 Plus (2026-04-02) | 57.9 |
| Qwen3.5 35B A3B | 57.1 |
| Opus 4.5 | 56.7 |
| Gemini 3 Pro Preview | 56.7 |
| Kimi K2.6 | 56.7 |
| Kimi K2.5 | 56.0 |
| Qwen3.5-9B | 55.9 |
| Qwen3-235B-A22B | 55.4 |
| GLM-5 | 55.1 |
| GPT-5.2 | 54.6 |
| GLM-5.1 | 54.6 |
| Qwen3.5-4B | 54.3 |
| Qwen3 Max Thinking (2026-01-23) | 54.2 |
| Qwen3-Next-80B-A3B-Thinking | 53.3 |
| DeepSeek-V4-Pro | 52.8 |
| Qwen3 30B A3B Thinking 2507 | 52.5 |
| GPT-5 Mini | 51.9 |
| gpt-oss-120b | 51.1 |
| gpt-oss-20b | 48.7 |
| Qwen3 4B Thinking 2507 | 47.1 |
| Qwen3.5-2B | 46.4 |
| Qwen3.5-0.8B | 42.4 |
| Qwen3 1.7B | 40.3 |