Nanbeige（BOSS 直聘）发布 Nanbeige4.3-33B 开源权重模型，Artificial Analysis 智能指数得分 20

Nanbeige has released Nanbeige4.3-33B, an open weights model matching Qwen3.8 27B (xhigh) at 20 on the Artificial Analysis Intelligence Index v5 with stronger terminal use but weaker knowledge

Nanbeige is the foundation model research team at BOSS Zhipin, the Chinese recruitment platform operated by Kanzhun Limited. Its dense 33B reasoning model is open weights under Apache 2.0, with text input and output only, a 524K-token context window, and is small enough to run locally on a laptop.

Key results:

➤ Competitive results at 30B total parameters. Nanbeige4.3-33B and Qwen3.8 27B (xhigh) both score 20, compared with 6 for Muse Glimmer (high, 30B).

➤ Terminal use and physics reasoning are relative strengths. Nanbeige scores 30% on Terminal-Bench 4.0 and 16% on CritPt v2's Hard Subset, vs. 6% and 4% respectively for Qwen3.8 27B.

➤ Different strengths to Qwen3.8 27B despite the same overall score. Nanbeige beats Qwen3.8 27B on AutomationBench-AA (53% vs. 48%), but trails on AA-Briefcase v1.1 (1162 vs. 1391 Elo), FrontierCode 2.0 (11% vs. 15%) and HLE-Diamond (7% vs. 15%).

➤ High abstention and low hallucination rates on Omniscience. Nanbeige scores -4 on AA-Omniscience v1.1 vs. -6 for Qwen3.8 27B, with a hallucination rate of 18% vs. 29%. It answers fewer questions correctly (12% vs. 17%) and declines 71% of questions.

➤ Similar intelligence with more output tokens per task. Nanbeige used 158K output tokens per Intelligence Index task vs. 110K for Qwen3.8 27B, about 44% more.

Additional model details:

➤ Type: Open weights reasoning (Apache 2.0 license)

➤ Size: 33B total parameters (dense)

➤ Modalities: Text input and output only

➤ Context window: 524K

➤ API availability: No confirmed public first- or third-party APIs

Nanbeige4.3-33B reaches 20 on the Intelligence Index with 33B total parameters alongside Qwen3.8 27B at 20 with 27B, and sets a new Pareto optimal point for Intelligence Index vs. Total Parameters. Both Nanbeige4.3-33B and Qwen3.8 27B are dense models. DeepSeek V4.1 Flash 1017 (max) scores 33 with 552B total parameters, but its MoE architecture activates only 16B per token.

Nanbeige4.3-33B's AA-Omniscience v1.1 result reflects selective answering. Its hallucination rate is 18% vs. 29% for Qwen3.8 27B, while its attempt rate is 29% vs. 43%. Qwen answers more questions correctly overall: 17% vs. 12%.

Nanbeige4.3-33B used 158K output tokens per Intelligence Index task, vs. 110K for Qwen3.8 27B - 44% more despite both scoring 20. This weighted average includes answer and reasoning tokens across the nine v5 evaluations, including agentic tool-use turns.

Full results across the Artificial Analysis Intelligence Index v5 evaluations