China’s Alibaba Group has released a Qwen3 LLM series said to be at the leading edge of open-source models, nearly achieving the performance of proprietary models from AI competitors OpenAI and Google. Alibaba says Qwen3 offers improvements in reasoning, tool use, instruction following and multilingual abilities. The Qwen3 series features eight new models — two that are mixture-of-experts and six built on dense neural networks. Their sizes range from 600 million to 235 billion parameters. The size and scope of the Alibaba slate maintains China’s accelerated AI pace in the wake of DeepSeek’s game-changing debut.

Alibaba says its Qwen3 models rival those of top U.S. competitors such as OpenAI’s o1 and o3-mini, xAI’s Grok 3 and Google’s Gemini 2.5 Pro, as well as China’s DeepSeek-R1.

“Overall, the benchmark data positions Qwen3-235B-A22B as one of the most powerful publicly available models, achieving parity or superiority relative to major industry offerings,” writes VentureBeat. Data provided by Alibaba indicates the flagship model “outperforms DeepSeek’s open source R1 and OpenAI’s proprietary o1 on key third-party benchmarks including ArenaHard (with 500 user questions in software engineering and math) and nears the performance of the new, proprietary Google Gemini 2.5-Pro.”

TechCrunch says all the new Qwen3 models “are — or soon will be — available for download under an open license on AI dev platform Hugging Face and GitHub.”

In a GitHub blog post Alibaba said it believes releasing the Qwen3 under the Apache 2.0 open-source license “will significantly advance the research and development of large foundation models,” helping the world’s global research and development community “build innovative solutions using these cutting-edge models.”

Due to Qwen3’s efficiencies, Alibaba says the series “significantly cuts deployment costs compared to other major models,” Bloomberg reports, noting that the company released a  Qwen2.5-Omni series “just a few weeks ago that can process text, pictures, audio and video — and is efficient enough to run directly on phones and laptops.”

The Qwen3 models utilize a hybrid reasoning approach “allowing users to toggle between fast, accurate responses and more time-consuming and compute-intensive reasoning steps (similar to OpenAI’s “o” series) for more difficult queries in science, math, engineering and other specialized fields,” VentureBeat explains.

“We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget,” the Alibaba developers write in the blog post. “This design enables users to configure task-specific budgets with greater ease.”