Several new artificial intelligence models have been announced, showcasing advancements in natural language processing and vision-language capabilities. LocalAI introduced multiple models including '72b-qwen2.5-kunou-v1', a successor to L3-70B-Euryale-v2.2, and 'deepthought-8b-llama-v0.01-alpha', which is built on LLaMA-3.1 with 8 billion parameters. Additionally, the 'l3.3-70b-euryale-v2.3' model has been released as a direct replacement for Euryale v2.2. SiliconFlowAI launched 'DeepSeek-V2.5-1210', an enhanced model for math, coding, and writing, now available on SiliconCloud. The new 'Llama-3.2-3B' model has joined the FlowerTune LLM leaderboard, while 'Llama 3.3', a 70 billion parameter model, offers comparable performance to larger models at a reduced cost. Cohere introduced 'Command R7B', the smallest and fastest model in their R series, designed for efficiency and lower deployment costs. DeepSeek also launched 'DeepSeek-VL2', a vision-language model available in sizes of 1.0B, 2.8B, and 4.5B parameters, utilizing a mixture of experts architecture for improved performance across various tasks. These developments reflect ongoing innovation in AI, with models increasingly focused on efficiency and versatility in applications.