Tencent · Hy4 (Hunyuan)
Tencent Hy4 preview
Tencent's open-weight (Apache-2.0) 770B/49B-active MoE text model with a 1M context, positioned for coding agents and office/productivity tasks; not yet benchmarked by Artificial Analysis.
- AA Intelligence
- —
- Input / output per 1M
- $0.834/ $2.501
- Output speed
- —
- Context window
- 1M tokens
Coding
- SWE-bench Pro#6 of 3165.7%
Leader: Claude Opus 5.5 at 89.9%
- Terminal-Bench 2.x#14 of 10385.4%
Leader: Claude Fable 5.1 at 91.4%
Reasoning and knowledge
- GPQA Diamond#24 of 14292.3%
Leader: GPT-6 Astra at 96.1%
Human preference
- LMArena WebDev#11 of 541631
Leader: Claude Opus 5.5 at 1818
Other published results
| Result | Score | Reported by | Source |
|---|---|---|---|
| DeepSWE | 64.3% | vendor | Hugging Face: tencent/Hy4-preview ↗ |
| Toolathlon Verified | 74.1% | vendor | Hugging Face: tencent/Hy4-preview ↗ |
| APEX-Agents | 37.1% | vendor | Hugging Face: tencent/Hy4-preview ↗ |
| Tencent internal expert eval (203 engineering tasks) | 2.99 score out of 4 | vendor | Tencent: Tencent releases and open-sources Tencent Hy4 preview ↗ |