Nanbeige

Chinese AI startup Nanbeige has released a new open-source model with 33 billion parameters, positioning it as a competitor to Alibaba's Qwen and Google's Gemma, with a focus on coding performance. The release adds to the growing roster of capable mid-sized open models, and observers are watching whether its coding benchmarks hold up against the established leaders.

Nanbeige, the AI team backed by Chinese recruitment platform BOSS Zhipin, has released Nanbeige4.3-33B as open source. The model is built for agentic coding tasks and supports context windows of up to 512,000 tokens, allowing it to process very large codebases in a single pass. The release adds to a growing wave of Chinese companies offering capable open-weight coding models.

Nanbeige has released Nanbeige4.3-33B as an open-source large language model, positioning it as a direct competitor to Qwen's 27B-class model on coding tasks. Early discussion centers on how a free, openly licensed 33-billion-parameter model can match or beat proprietary-leaning alternatives, with developers weighing benchmarks, licensing terms and practical use in code generation workflows.