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

Nanbeige4.3-33B is a 33 billion parameter model with a 524,288-token context window. Its model card compares it with Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B, all evaluated at their highest thinking effort. On Terminal-Bench 2.1, the card lists 80.9 for Nanbeige and 73.0 for the closest comparison model in its table. The release comes from a Chinese lab backed by the recruitment platform BOSS Zhipin. It lands in the same week as other open-weight releases from Chinese labs, including the Qwen3.8-27B family and Moonshot's Kimi K3.1 Pro, both already covered on this site. For teams choosing open weights, a model in this size class may fit on local hardware. Independent benchmark runs should come before any move away from a hosted API, since the scores so far come from the developer's own evaluation.