Nanbeige Releases Open 33B Model Claiming Wins Over Qwen3.8-27B

Nanbeige4.3-33B supports a 524,288-token context window and is published on Hugging Face. The model card reports 80.9 on Terminal-Bench 2.1 against 73.0 for Qwen3.8-27B, and 23.8 on Terminal-Bench 4.0 against 6.3. The authors compare against Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B, all run at their highest thinking effort. The card also shows gaps, since Qwen3.8-27B scores higher on SWE-Bench Pro. The release comes from a lab backed by recruitment platform BOSS Zhipin. A secondary summary says the weights ship under Apache 2.0 with reasoning on by default, which should be checked against the model card before relying on it. For teams choosing open weights, the release adds another roughly 30B-class option for agentic coding and long-context work. Vendor benchmark tables are self-run, so evaluate on your own harness and tasks before swapping a production model.