| Mobile Web

Search results for multimodal AI

AI & Enterprise

Hands-on with Alibaba\'s Qwen3.8-27B: strong performance but thinks too much

Alibaba has open-sourced Qwen3.8-27B, a multimodal AI model sized to run locally on higher-spec MacBook Pros or Mac Studios. Simon Willison, who developed Datasette and co-developed Django, published a hands-on review. He said the model’s performance is impressive but criticised its default “xhigh” deep-thinking setting for taking too long on simple tasks. He cited a 21-minute image-generation response and noted slower speeds even on high-end hardware.

AI & Enterprise

Small AI models near large-model performance as Dokpamo second evaluation nears

Small AI models that can be installed on PCs are evolving rapidly, raising attention over whether they can rival large models that run on high-capacity computers. Tech companies are releasing local-run models aimed at cost and privacy needs. The government is set to disclose second-round results for its independent AI foundation model project and narrow four teams to three. The article also lists recent moves by Korean and global companies on AI models, tools, pricing and partnerships.

AI & Enterprise

Alibaba releases open-source AI model that can run on MacBook

Alibaba has released an open-source multimodal AI model, Qwen3.8-27B, optimised for local use, including its weights under an Apache 2.0 licence. The company also plans to publish weights for its top model, Qwen3.8-2.4T-A95B, letting developers choose between lightweight local models and frontier-grade agent tasks. Alibaba says Qwen3.8-27B can run on high-end MacBook Pro or Mac Studio systems and supports up to 1,000,000 tokens with YaRN.