KataGo-Bench-1K: leaderboard
Go next-move prediction benchmark with 1,000 19x19 board positions sampled from KataGo annotation data. Models read the move history and rendered board state, then output the next move; scoring is accuracy against KataGo candidate moves.
Metric: Next-move Accuracy (%). Source: github.com. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 34.3 | #249 |
| 2 | O1 Mini | 27.3 | #323 |
| 3 | DeepSeek R1 | 17.6 | #251 |
| 4 | Qwen 2.5 7B Instruct | 8 | #862 |
| 5 | Qwen 2.5 32B Instruct | 6.8 | #498 |
| 6 | DeepSeek R1 Distill Qwen 32B | 4.7 | #655 |
| 7 | Qwen 2.5 32B | 1.5 | #546 |
| 8 | Qwen 2.5 7B | 1.4 | #922 |
| 9 | DeepSeek-R1-Distill-Qwen-7B | 0.6 | #1415 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=katago-bench-1k · How It Works · Data refreshed daily, snapshot 2026-10-20.