KOR-Bench (Knowledge-Orthogonal Reasoning benchmark) — Real published reasoning benchmark; 'KOR-Bench'/'KORBench' spelling variants pooled correctly.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Seed Oss 36B Instruct | ByteDance | 70.6 | 1 | 2026-06-15 |
| 2 | Qwen3 30B A3B | Alibaba | 70.2 | 1 | 2026-06-15 |
| 3 | DeepSeek-R1-Distill-Qwen-32B | DeepSeek | 66.6 | 2 | 2026-06-05 |
| 4 | Qwen3 32B | Alibaba | 65.4 | 1 | 2026-06-15 |
| 5 | SynLogic Mix 3 32B | MiniMax | 65 | 1 | 2026-06-05 |
| 6 | SynLogic 32B | MiniMax | 62.2 | 1 | 2026-06-05 |
| 7 | Qwen2.5 Instruct 32B | Alibaba | 54.7 | 1 | 2026-06-05 |
| 8 | SynLogic 7B | MiniMax | 48.1 | 1 | 2026-06-05 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.