BOSS Zhipin launches the 33B agent model, leading in 17 out of 20 self-evaluation metrics.

ME News reports that on October 26 (UTC+8), Nanbeige Lab, a subsidiary of the recruitment platform BOSS Zhipin, released the open-source model Nanbeige4.3-33B, designed for programming and office agents, supporting a 512K context length under the Apache 2.0 license. The official comparison with Qwen3.8-27B, Meta Muse Glimmer 30B, and Gemma 4 31B shows that Nanbeige4.3-33B ranked first in 17 out of 20 published tests. It achieved a Terminal-Bench 2.1 score of 80.9, surpassing Qwen’s 73.0; DeepSWE reached 65.5 compared to a baseline of 42.5; and on the office automation benchmark Automation Bench, it scored 54.2, higher than Qwen’s 38.3. However, Qwen still leads in SWE-Bench Pro and GDPval. Previously, Nanbeige primarily developed 3B-scale small models; the previous generation, Nanbeige4.2-3B, improved effective computational depth without increasing parameters by reusing Transformer layers. This time, the team first trained the 33B model to serve as a teacher, then plans to distill programming and tool-use capabilities into the upcoming Nanbeige4.3-3B. (Source: ChainCatcher)