The 128K-context MoE flagship that introduced sparse attention
Added Dec 1, 2025
DeepSeek V3.2 is a 128K-context mixture-of-experts model that unified thinking and non-thinking modes in the V3 line and introduced DeepSeek Sparse Attention. It served as the December 2025 flagship before V4 and remains available for self-hosting and as a historical comparison point.
Why: V3.2 introduced DeepSeek Sparse Attention and unified thinking modes, making it the architectural bridge that enabled the later 1M-context V4 family.
Strong general-reasoning model with interleaved thinking
Added Dec 1, 2025
GLM-4.7 is Z.ai's general-reasoning tier, offering a 200K context window and 128K maximum output. It is positioned between the mid-range GLM-4.6 and the GLM-5 flagships, with emphasis on interleaved thinking and broad tool-use tasks.
Why: GLM-4.7 is the cost-effective sweet spot for long-context reasoning and general-purpose agent work before stepping up to the GLM-5 series.
Vision-language model for visual reasoning and UI replication
Added Dec 1, 2025
GLM-4.6V is a 2025 vision-language model in the GLM family, offering visual reasoning, tool calling, and frontend code replication. It supports a 128K context window and a 32K maximum output.
Why: GLM-4.6V is the practical vision tier for turning screenshots and images into working code or structured analysis.
Tripo 4.0 is the latest generation of Tripo AI's text- and image-to-3D pipeline, emphasizing high-fidelity geometry, detailed textures, PBR materials, and game-engine-ready topology.
Essential AI, whose CEO co-wrote Google's Attention Is All You Need paper, unveils Rnj-1, an 8B-parameter open model with SWE-bench performance close to GPT-4o. The long-term advancement and equitable diffusion of AI technologies crucially depend on their deve…
Mistral AI, Europe's leading artificial intelligence startup, released its next generation of models pitched as more adaptable than alternatives from larger rivals OpenAI and Alphabet Inc.'s Google.
China's DeepSeek unveiled two new versions of an experimental artificial-intelligence model it released weeks ago, adding fresh capabilities the startup said would help with combining reasoning and executing certain actions autonomously.