{"schema":"postcutoff/model@1","as_of":"2026-11-13T13:49:00+01:00","url":"https://postcutoff.com/m/nanbeige4-3-33b/","md":"https://postcutoff.com/m/nanbeige4-3-33b/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"nanbeige4-3-33b","name":"Nanbeige4.3-33B","org":"Nanbeige (Kanzhun / BOSS Zhipin)","family":"Nanbeige4.3","released":"2026-10-24","status":"current","type":"reasoning-llm","modality_in":["text"],"modality_out":["text"],"open_weights":true,"model_license":"apache-2.0","context_window":524288,"max_output":262144,"knowledge_cutoff":null,"pricing":null,"price_line":"Price not published","access":[{"provider":"Hugging Face","model_id":"Nanbeige/Nanbeige4.3-33B","url":"https://huggingface.co/Nanbeige/Nanbeige4.3-33B"},{"provider":"Hugging Face (base model)","model_id":"Nanbeige/Nanbeige4.3-33B-Base","url":"https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base"}],"capabilities":[{"name":"512K context in a ~33B open model for coding agents","detail":"Supports up to 524,288 tokens; chat template has enable_thinking and preserve_thinking (keep earlier reasoning across turns, recommended for multi-turn tool use and code agents); XML (recommended) or JSON tool calls.","first":false,"discovered":"launch","source":"https://huggingface.co/Nanbeige/Nanbeige4.3-33B"},{"name":"Teacher model for small-agent distillation","detail":"Built as the flagship teacher of the Nanbeige4.3 series, whose agent skills are to be distilled into smaller models such as a planned Nanbeige4.3-3B.","first":false,"discovered":"launch","source":"https://huggingface.co/Nanbeige/Nanbeige4.3-33B"}],"entry":"2026-10-24-nanbeige4-3-33b-open-weights","notes":"~33.1B parameters (safetensors). Self-reported to lead Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B on most agentic coding and office benchmarks (Terminal-Bench 2.1/4.0, SWE-Atlas-QnA, NL2Repo, DeepSWE, GDPval, Tau3-Bench…); not independently verified. Recommended: temperature 1.0, top_p 0.95, up to 65,536 new tokens for agentic tasks; 0.6 and up to 262,144 for reasoning/chat. Runs on Transformers, SGLang, vLLM (Qwen3-style reasoning and tool parsers). No hosted API found.","verified":"2026-10-26","body_md":"Open-weight (Apache 2.0) agentic coding model from the Nanbeige team at Kanzhun, the company behind China's BOSS Zhipin recruitment app.\n\n```bash\nvllm serve Nanbeige/Nanbeige4.3-33B --max-model-len 524288\n```","page_url":"https://postcutoff.com/m/nanbeige4-3-33b/","events_after":null,"major_after":null,"historic_after":null,"missing_at_launch":null,"events_since_release":436,"you_url":"https://postcutoff.com/you/nanbeige4-3-33b/","briefings":null}