Alibaba’s AI research unit Qwen·Tongyi Lab has unveiled Qwen3.5-Omni, an omni-modal model covering text, image, audio and video understanding as well as speech generation. The lab highlighted real-time responses and long-input processing, with a maximum sequence length of 256,000 and support for up to 10 hours of audio input. It also disclosed training data and a dual mixture-of-experts structure, and introduced three product variants with offline and real-time APIs.

Google\'s Gemini 3.1 Pro is a major upgrade from 3.0, improving its ability to carry out specific tasks beyond simple Q&A. A report introduced five prompts to draw out its strengths: reverse-checking plans for weak points, matching video aesthetics to real places, analysing images to help solve problems, generating interactive SVG animations, and using Deep Research to produce practical guides. Results can vary with prompt design and verification.