LinkTencent Hy TeamAnnouncement
Hy4-preview marks a notable step forward for open-weights frontier models, combining very large parameter counts with sparse activation to keep serving viable. Its permissive Apache 2.0 license and turn-key deployment paths in vLLM and SGLang allow teams to evaluate large-scale reasoning and coding capabilities within their own infrastructure.
Tencent incorporates key recent architectural designs—including Gated Sparse Attention and multi-token prediction—to deliver efficiency gains across 78 layers and 1M tokens of context. Rather than focusing purely on synthetic evaluations, the team highlights productivity gains across software development and workplace automation. While the release shows strong metrics against competing models such as GLM 5.3 and Kimi K3, Tencent explicitly acknowledges that reasoning efficiency and over-thinking remain active targets for follow-up iterations.
Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.
Similar in content, found by the AI search.
Repository:Tencent-Hunyuan/Hy4-preview
The Tencent Hy Team has introduced Hy4-preview, a new flagship Mixture-of-Experts (MoE) model featuring 770B total parameters with 49B activated per token. The architecture incorporates MoE routing, Gated DeepSeek Sparse Attention, and a native MTP layer for speculative decoding.
344starsPython
Link:deepseek-ai
DeepSeek has published the model card and weights for DeepSeek-V4-Pro-DSpark on Hugging Face. The release equips the DeepSeek-V4-Pro MoE model with an attached speculative decoding module (DSpark) for accelerated inference.
Link:deepseek-ai
DeepSeek-AI has officially released DeepSeek-V4-Flash-0731 on Hugging Face. The 304B-parameter model supersedes the preview version, featuring integrated DSpark speculative decoding and significantly enhanced agentic performance.
Article:DeepSeek AI
DeepSeek has open-sourced the preview release of DeepSeek-V4. The line-up features DeepSeek-V4-Pro and DeepSeek-V4-Flash, standardising a 1M context window and introducing architectural innovations to cut compute and memory overhead.
Link:Xiaomi MiMo Team
Xiaomi has unveiled MiMo-V2.5-Pro, an open-source Mixture-of-Experts model featuring 1.02 trillion total parameters with 42 billion active. The model supports a 1M-token context window and is specifically designed for complex agentic workflows and long-horizon software engineering.
Article:Simon Willison
DeepSeek has introduced DeepSeek-V4-Pro and DeepSeek-V4-Flash under an MIT licence. Simon Willison reviews their Mixture of Experts architecture, KV cache efficiency gains, and pricing compared to OpenAI, Google, and Anthropic.
Parts of this content were created with the assistance of AI.
