Alibaba Cloud has launched Qwen3, the latest iteration in the Qwen series of large language models (LLMs), which reportedly outperforms leading models such as DeepSeek-R1, o1, and Gemini-2.5-Pro. Qwen3 features 32 billion parameters and supports hybrid reasoning and multilingual capabilities, with applications in solving STEM problems and executing agentic actions. The model is integrated with Lambda's Inference API for broader accessibility. Concurrently, LocalAI has introduced several new models, including Qwen2.5-Omni-7B, a multimodal model capable of processing text, image, and audio inputs, and the pku-ds-lab_fairyr1 series (14B-preview and 32B), which are efficient LLMs built on DeepSeek-R1 architecture. Additionally, LocalAI released Moondream2, a compact vision-language model. The DeepSeek-R1-0528 model has also been made available with a promotional discount for inference services.