AI · SEMICONDUCTOR · FACT CHECK
SKD WIREMostly True
Chinese AI startup Nanbeige has publicly released Nanbeige4.3-33B, a 33B-parameter open-source large language model (LLM), and the model posted stronger results than Qwen's 3.8-27B on coding benchmarks, according to materials published by the company.
Nanbeige made the model available through its official Hugging Face repository (Nanbeige/Nanbeige4.3-33B), where it also published benchmark figures for the new release. The release follows the company's practice of distributing its models in open-source form, allowing external developers to download and run the weights directly.
The company announced the release via its official X (formerly Twitter) account, and the announcement text was accessible only through that post and articles embedding it.
According to the benchmark figures Nanbeige published alongside the release, Nanbeige4.3-33B scored ahead of Qwen's 3.8-27B model on coding benchmarks. The comparison places the 33B-parameter model against a smaller 27B-class competitor, positioning Nanbeige's release as competitive in the mid-to-large open-source model segment.
The published figures cover coding performance among other evaluation areas; the full benchmark table was released as images on the Hugging Face page.
The claim that Nanbeige4.3-33B outperforms Qwen on coding rests on benchmark data published by Nanbeige itself, and the official announcement was confirmed to exist only via the company's X post and embedded citations. Independent benchmark comparisons of the two models were not part of the material reviewed, and the company's self-reported figures should be read with that in mind.
Nanbeige's release of the 33B open-source model and its reported benchmark lead over Qwen's 27B-class model are largely supported by the company's own published materials, though the performance comparison comes from self-reported data — making the overall claim Mostly True.
