Tencent

Tencent is the Shenzhen company behind Weixin (WeChat) and some of the most popular games in the world, and its Hy team, which publishes under the Tencent-Hunyuan organization on GitHub, has become one of the more aggressive open-weight publishers in China. On August 28, 2026 Tencent released and open sourced Hy4 preview, a mixture-of-experts with 770 billion total parameters and 49 billion active per token, a 1M token context window, and weights under plain Apache 2.0 on Hugging Face, ModelScope, GitCode, and CNB. API access runs through Tencent Cloud TokenHub and OpenRouter at $0.834 per million input tokens, $2.501 output, and $0.042 for cache hits, and the model is built into Tencent products including WorkBuddy, CodeBuddy, Yuanbao, and ima. The architecture borrows from DeepSeek and GLM, using gated DeepSeek Sparse Attention with IndexCache, and adds a 10 billion parameter multi-token prediction layer for speculative decoding. The evidence is thinner than the spec sheet. The headline result is an internal blind evaluation in which 163 Tencent experts scored Hy4 preview 2.99 out of 4 on 203 engineering tasks, a hair ahead of Kimi K3 at 2.94 and GLM-5.3 at 2.92, and the benchmark appendix ships as images rather than a table. Tencent says Hy4 preview also helped optimize its own training and inference stack, lifting throughput 31.8 percent, and that more Hy4 models are coming soon. The preview-first pattern is deliberate: Tencent shipped Hy3 preview the same way and iterated on user feedback before the full release.