Hunyuan-4B-Instruct: benchmark results
Tencent Hunyuan dense instruction checkpoint with grouped-query attention and switchable fast and slow thinking. The publisher supports a 256K context; pretraining checkpoints and quantized releases remain separate. Provider: Tencent. Released 2025-07-30. Access: Open.
Aggregate Rating 1324 ± 29, rank #8260 of 10916 rated models, from 265 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Portuguese NLU - ScaLA PT | 26.53 | Matthews correlation coefficient (%) | 80.9 |
| EuroEval French NLU - ScaLA FR | 41.84 | Matthews correlation coefficient (%) | 76.1 |
| EuroEval Polish Summarization - PSC | 23.16 | chrF3++ (%) | 73.5 |
| EuroEval Italian NLU - ScaLA IT | 27.18 | Matthews correlation coefficient (%) | 70.9 |
| EuroEval English NLU - CoNLL EN | 74.41 | Micro F1 excluding MISC (%) | 63.6 |
| EuroEval Bulgarian NLU - Cinexio | 34.76 | Matthews correlation coefficient (%) | 63.5 |
| EuroEval Spanish NLU - CoNLL ES | 63.93 | Micro F1 excluding MISC (%) | 61 |
| EuroEval Spanish NLU - ScaLA ES | 21.23 | Matthews correlation coefficient (%) | 60.3 |
| EuroEval English NLU - ScaLA EN | 46.2 | Matthews correlation coefficient (%) | 57.1 |
| EuroEval Faroese NLU - ScaLA FO | 4.42 | Matthews correlation coefficient (%) | 57 |
| EuroEval Catalan NLU - ScaLA CA | 8.93 | Matthews correlation coefficient (%) | 55 |
| EuroEval Dutch NLU - CoNLL NL | 58.5 | Micro F1 excluding MISC (%) | 55 |
Interactive version: theaggregate.ai/model?slug=hunyuan-4b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.