Alibaba releases Qwen3 models, flooding the market

Six new Qwen3 models covering text, vision, audio, and safety were just announced by Alibaba this week. The most notable of these is the recently revealed 1T parameter Qwen-Max. The specifics: While its Heavy version scores flawlessly on all numerical reasoning criteria, Max has near-frontier capabilities in coding and agentic tasks. In addition to providing speech interpretation in 19 languages and generation in 10 languages, Omni can analyze text, graphics, audio, and video. On a number of benchmarks, VL outperforms the best closed models and is ranked as the best non-reasoning, open-source visual model. Additionally, Alibaba provided Guard models for safety moderation, new improved Coder variations, and LiveTranslate-Flash for real-time interpretation. We've been conditioned by OAI, Anthropic, and Google to anticipate a slower release timetable, but Chinese behemoths like Alibaba are adopting the opposite strategy, flooding the market with near-frontier, specialized options at a rapid-fire pace. Since DeepSeek's R1, Qwen believes this is the closest a Chinese lab has been to the frontier.