Hy3-preview (Reasoning): benchmark results
Tencent Hunyuan Hy3 preview (295B A21B open MoE) evaluated with reasoning enabled. Provider: Tencent. Released 2026-04-23. Access: Open.
Aggregate Rating 1674 ± 17, rank #442 of 2168 rated models, from 62 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| VEHBench - Corrupted-State Recovery | 47.4 | P3 success (%; share of corrupted or misleading design traje | 100 |
| LLMEval-Logic Base | 75.3 | Accuracy (%) | 92.3 |
| AA Omniscience - Software Engineering (SWE) - R | 30 | Accuracy (%) | 88.7 |
| UGI - Natural Intelligence | 50.11 | NatInt Score | 88.7 |
| AA Omniscience - Software Engineering (SWE) - JavaScript | 46.36 | Accuracy (%) | 88.4 |
| AA TAU-2 Bench | 92.69 | Accuracy (%) | 88.3 |
| Tau2-Bench Telecom | 92.69 | Success Rate (%) | 88.1 |
| UGI - Writing | 51.12 | Writing Score | 87.1 |
| AA Omniscience - Software Engineering (SWE) - Swift | 52 | Accuracy (%) | 86.7 |
| AA Omniscience - Software Engineering (SWE) - Rust | 64 | Accuracy (%) | 86.5 |
| AA Omniscience - Software Engineering (SWE) - Dart | 30 | Accuracy (%) | 85.2 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 30 | Accuracy (%) | 84.1 |
Interactive version: theaggregate.ai/model?slug=hy3-preview-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.