Nanbeige publishes Apache 2.0 Nanbeige4.3-33B coding model

Nanbeige has released open 33B weights it says lead most of its own ~30B coding and cowork benchmarks, with a 3B distill planned.

Nanbeige LLM Lab, inside recruitment platform BOSS Zhipin, published Nanbeige4.3-33B and a base checkpoint on Hugging Face under Apache 2.0. The dense 33B Llama-style model is the flagship teacher for the Nanbeige4.3 series and targets agentic coding and professional cowork tasks, with a 524,288-token window and thinking on by default. The lab says it leads Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B on 17 of 20 self-run rows, including Terminal-Bench 2.1 at 80.9, while trailing Qwen on SWE-Bench Pro at 60.3 versus 61.7. MarkTechPost says several competitor figures conflict with those labs’ own model cards. A Nanbeige4.3-3B distillation is described as upcoming.