Hy3-preview (High): benchmark results
Tencent Hunyuan Hy3 preview (295B A21B open MoE) evaluated at the high reasoning-effort setting. Provider: Tencent. Released 2026-04-23. Access: Open.
Aggregate Rating 1687 ± 11, rank #396 of 2168 rated models, from 82 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CLBench Life - Personal Information Fragments | 20 | Task solving rate (%) on the 45 personal-information-fragmen | 100 |
| OpenCompass Math - College (CompassBench 2604) | 81.3 | Score (%) | 90.9 |
| SafePyramid - L1 Rule Dependencies | 55.9 | Rule matching rate (%) on the 1,000 L1 cases (policies that | 90.9 |
| CLBench Life - Fragmented Information & Revisions | 15.9 | Solving Rate (%) | 89.3 |
| OpenCompass Language - Instruction Following (CompassBench 2604) | 76 | Score (%) | 86.4 |
| OpenCompass LLM - Math (CompassBench 2604) | 74.5 | Score (%) | 81.8 |
| OpenCompass Language - Creation (CompassBench 2604) | 75.4 | Score (%) | 81.8 |
| OpenCompass Math - College | 81 | Score (%) | 78.6 |
| CLBench Life - Creation & Revision Histories | 14.4 | Task solving rate (%) on the 45 creation and revision histor | 77.8 |
| CLBench Life | 15.7 | Solving Rate (%) | 75 |
| OpenCompass LLM - Language (CompassBench 2604) | 74.4 | Score (%) | 75 |
| Buyout Game (Lechmazur) | 1620.9 | Bradley-Terry Rating | 74.3 |
Interactive version: theaggregate.ai/model?slug=hy3-preview-high · How It Works · Data refreshed daily, snapshot 2026-10-20.