Hunyuan-T1-20250822: benchmark results
Tencent's Hunyuan-T1 (August 22, 2025 checkpoint), a Hybrid-Transformer-Mamba Mixture-of-Experts reasoning model activating 52 billion parameters via dynamic expert routing, built on the TurboS foundation. Provider: Tencent. Released 2025-08-22. Access: API.
Aggregate Rating 1656 ± 27, rank #296 of 1635 rated models, from 68 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenCompass Reasoning - Common - Chinese (CompassBench 2510) | 84.9 | Score (%) | 100 |
| OpenCompass Knowledge - Humanities - Chinese (CompassBench 2510) | 92 | Score (%) | 98.1 |
| OpenCompass Math - College - Chinese (CompassBench 2510) | 56 | Score (%) | 86.5 |
| OpenCompass Code - Competition - Chinese (CompassBench 2510) | 65.2 | Score (%) | 84.6 |
| OpenCompass Language - Instruction Following - Chinese (CompassBench 2510) | 81 | Score (%) | 84.6 |
| SuperCLUE General (September 2025) - Hallucination Control | 77.96 | Score | 83.3 |
| SuperCLUE-Fin (November 2025) - Overall | 75.71 | Score | 83.3 |
| SuperCLUE-Fin (October 2025) - Overall | 75.78 | Score | 83.3 |
| OpenCompass LLM - Code (CompassBench 2510) | 61.6 | Score (%) | 82.7 |
| OpenCompass Language - Dialogue - Chinese (CompassBench 2510) | 94.6 | Score (%) | 82.7 |
| OpenCompass LLM - Language - Chinese (CompassBench 2510) | 86.5 | Score (%) | 76.9 |
| OpenCompass Language - Creation - Chinese (CompassBench 2510) | 91.3 | Score (%) | 76.9 |
Interactive version: theaggregate.ai/model?slug=hunyuan-t1-20250822 · How It Works · Data refreshed daily, snapshot 2026-10-15.