KINA — leaderboard
Metric: Accuracy (%). Source: arxiv.org. 40 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 53.17 |
| 2 | Claude Opus 4.6 | 49.92 |
| 3 | GPT-5.4 | 48.55 |
| 4 | Gemini 3 Flash (Preview) | 43.91 |
| 5 | Qwen 3.5 397B A17B | 42.99 |
| 6 | Kimi K2.5 | 40.24 |
| 7 | GPT-5.2 | 39.52 |
| 8 | Qwen 3.5 27B | 39.35 |
| 9 | Qwen 3.5 122B A10B | 38.88 |
| 10 | DeepSeek V3.2 (Thinking) | 38.01 |
| 11 | Qwen 3 Max (2025-09-23) | 35.9 |
| 12 | GLM-5 | 35.85 |
| 13 | Qwen 3.5 35B A3B | 35.43 |
| 14 | Grok 4.1 Fast (Reasoning) | 33.73 |
| 15 | Qwen 3 235B A22B (Thinking) | 32.15 |
Interactive version: aibenchmarks.dev/benchmark?slug=kina · How the rankings work · Data refreshed daily, snapshot 2026-07-06.