
09/25/2025
Alibaba just released a barrage of new Qwen3 models this week, dropping six new variants across text, vision, audio, and safety — highlighted by the newly unveiled 1T parameter Qwen-Max.
The details:
Qwen3-Max shows near-frontier capabilities in coding and agentic tasks, while its Heavy version achieves perfect scores across math reasoning benchmarks.
Qwen3-Omni is capable of processing text, images, audio, and video, while supporting speech understanding in 19 languages and generation in 10 languages.
Qwen3-VL grades out as the top non-reasoning and open-source visual model, while also surpassing top closed models on a series of benchmarks.
Qwen3-Alibaba also released LiveTranslate-Flash for real-time interpretation, Guard models for safety moderation, and new upgraded Coder variants.
Source: The Rundown AI Newsletter