LinkXiaomi MiMo TeamAnnouncement
Open-source frontier models continue to narrow the gap with proprietary industry leaders. MiMo-V2.5-Pro demonstrates that algorithmic efficiency, such as hybrid attention and multi-token prediction, allows models to maintain coherent multi-hour agent trajectories and deliver complex software engineering outputs without exponential token overhead.
Xiaomi focuses heavily on long-horizon coherence and token economics. By combining Sliding Window Attention with Global Attention, the architecture significantly reduces the KV-cache footprint across long contexts. The three-stage post-training regimen leverages domain-specific reinforcement learning teachers unified through Multi-Teacher On-Policy Distillation (MOPD). As demonstrated by its integration into agent frameworks like Claude Code, MiMo-V2.5-Pro provides a viable high-performance open-weights option for iterative, tool-heavy engineering tasks.
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.
Link:Moonshot AI
Moonshot AI has released Kimi K3, an open-weight native multimodal agentic model featuring 2.8 trillion total parameters and a 1-million-token context window. The architecture uses a sparse MoE setup that activates 104 billion parameters per token.
Link:Tencent Hy Team
The Tencent Hy Team has released Hy4-preview, an open-source Mixture-of-Experts (MoE) model featuring 770B total parameters and 49B activated parameters per token. The flagship model supports a 1M token context window and is tuned for software engineering, office analytics, and scientific research.
Link:Z.ai
Z.ai has released GLM-5.2 on Hugging Face under the permissive MIT licence. The flagship open-weights model is designed for long-horizon agentic and coding workflows, featuring a robust 1M-token context window. Architectural updates including IndexShare and an enhanced MTP layer significantly cut computation FLOPs and accelerate speculative decoding.
Link:ornith-ai
ornith-ai has released Ornith-1.5-397B, an open-source multimodal mixture-of-experts model under the MIT licence. With 397 billion parameters and an end-to-end self-improvement loop, it delivers coding and reasoning benchmark performance comparable to top proprietary models.
Link:Apodex AI
Apodex AI has released Apodex-1.1-mini under the Apache-2.0 licence on Hugging Face. Based on the qwen3_5_moe architecture, the model is designed for long-horizon research and agentic workflows with native tool execution.
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.
Parts of this content were created with the assistance of AI.
