MODELS
混元 Hunyuan Large
Tencent's open-source 389B MoE flagship — biggest open-source MoE at release.
Specs
- Context window
- 256,000
- Max output
- 4,096
- Modalities
- text
- Tool use
- ✓
- Vision
- —
- Streaming
- ✓
- License
- tencent-hunyuan-community
- Released
- 2024-11-05
Pricing
- Input / 1M
- $0.60
- Output / 1M
- $1.80
Cost estimate
Hunyuan Large (November 2024) is Tencent's open-source 389B-parameter mixture-of-experts model — at release the largest open-source MoE on the market with 52B activated per token. 256K context, strong on Chinese (Mandarin) tasks. Available as weights on Hugging Face under Tencent Hunyuan Community Licence (commercial use allowed below 100M MAU) and via Tencent Cloud API at PRC-competitive pricing.
Editor's verdict
Strong on benchmarks, but real-world adoption is thinner than DeepSeek V3 because the licence has more strings (the 100M-MAU clause spooks legal teams) and self-hosting 389B params demands an 8×H100 minimum. Use Tencent Cloud API if you're already on WeChat/QQ ecosystems. For most builders DeepSeek V3 is the more practical open-source MoE pick.
Reviews
No reviews yet. Be the first.
Last updated: 2026-04-29