Hy3: benchmark results
Tencent's Hunyuan 3 open MoE model (295B total, 21B active), tuned for agentic tool calling and reasoning. Provider: Tencent. Released 2026-07-06. Access: Open.
Aggregate Rating 1768 ± 7, rank #93 of 1656 rated models, from 402 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA Omniscience - Software Engineering (SWE) - Swift | 76 | Accuracy (%) | 99.3 |
| ObviousBench | 100 | Answer pass³ (%) | 97.7 |
| AA Omniscience - Software Engineering (SWE) - PHP | 58 | Accuracy (%) | 96.8 |
| AGI-Eval Community - Subject Reasoning | 88.92 | Accuracy (%) | 96.8 |
| AA Omniscience - Software Engineering (SWE) - Java | 37 | Accuracy (%) | 96.4 |
| AGI-Eval Community - Subject Knowledge | 89.91 | Accuracy (%) | 96.4 |
| AA Omniscience - Software Engineering (SWE) - Python | 46.5 | Accuracy (%) | 96.2 |
| AGI-Eval Community - Subject Reasoning (Chinese) | 93.24 | Accuracy (%) | 96 |
| Enkrypt AI - Risk Score | 15.15 | Risk Score | 95.9 |
| AA Omniscience - Software Engineering (SWE) - HTML | 64 | Accuracy (%) | 95.7 |
| AA Omniscience - Software Engineering (SWE) - TypeScript | 53.33 | Accuracy (%) | 95.2 |
| AA Omniscience - Software Engineering (SWE) - Rust | 74 | Accuracy (%) | 95.1 |
Interactive version: theaggregate.ai/model?slug=hy3 · How It Works · Data refreshed daily, snapshot 2026-10-20.