AnswersPublished18 sources
Nanbeige4.3-33B targets coding agents and professional office work
Announced on October 26, 2026, Nanbeige4.3 33B is a 33.1B open weight model for coding agents and professional “cowork” tasks, listed under Apache 2.0. Nanbeige also framed the 33B release as a teacher for smaller models, beginning with a planned Nanbeige4.3 3B.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Announced on October 26, 2026, Nanbeige4.3 33B is a 33.1B open weight model for coding agents and professional “cowork” tasks, listed under Apache 2.0.
- Nanbeige also framed the 33B release as a teacher for smaller models, beginning with a planned Nanbeige4.3 3B.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did BOSS Zhipin parent Kanzhun’s Nanbeige team announce in its October 26, 2026 release of Nanbeige4.3-33B, including its availability. Article summary: Nanbeige, the model team within BOSS Zhipin parent Kanzhun, announced Nanbeige4.3-33B on October 26, 2026 as an open-weight model focused on coding agents and professional office “cowork” tasks, with weights available th. Topic tags: general, general web. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers, click
Nanbeige, the model team within BOSS Zhipin parent Kanzhun, announced Nanbeige4.3-33B on October 26, 2026. The 33.1-billion-parameter BF16 model is aimed at coding agents and professional office “cowork” tasks, and its weights are available on Hugging Face under Apache 2.0. The team also described it as a teacher model for distilling agent capabilities into smaller models. 1
7
14
What Nanbeige released
The announcement covered Nanbeige4.3-33B and a companion base model, Nanbeige4.3-33B-Base. Reporting says both Hugging Face repository pages were created on October 24, ahead of the October 26 announcement. 1
Nanbeige’s positioning is agent-first: the model is presented for coding and office workflows that use tools, not just ordinary chat. The team described the 33B model as the first teacher in a plan to distill its agent capabilities into smaller models, with Nanbeige4.3-3B named as the first planned student. 1
What the benchmark results show—and don’t
Nanbeige’s model card compares the model with Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B. It says the models were tested at their highest thinking effort, using the same inference settings and evaluation harness for each benchmark. On Terminal-Bench 2.1, the card reports 80.9 for Nanbeige4.3-33B and 73.0 for Qwen3.8-27B. 17
Those figures are the publisher’s own reported results, not independent replication. The shared test conditions described in the card make the comparison more interpretable, but do not by themselves establish that the results have been independently verified. 1
17
Long context and agent controls
The model is listed with a maximum context length of 524,288 tokens, or 512K. Its chat template includes two relevant controls: enable_thinking, which turns reasoning generation for the current response on or off and is enabled by default, and preserve_thinking, which determines whether reasoning from earlier assistant turns is retained. Nanbeige recommends disabling the latter for general chat and question answering, and enabling it for multi-turn tool use, office tasks and coding-agent workflows. 3
9
The model supports XML and JSON formats for tool calls; available model guidance recommends XML. 28
31 These options support the model’s agent-workflow positioning, but they are configuration features—not evidence on their own of better performance in real deployments.
Availability and practical caveat
The release was listed under the Apache 2.0 license, and the official Nanbeige model repository is on Hugging Face. Anyone considering use should check the repository’s license file and model documentation for the applicable terms and implementation details. 7
9
14
The clearest takeaway is the release’s combination of a long context window, tool-use options and a teacher-to-smaller-model plan, all aimed at agentic coding and professional work. Its benchmark claims are worth treating as the team’s reported results until independently reproduced.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Nanbeige4.3-33B targets coding agents and professional office work"?
Announced on October 26, 2026, Nanbeige4.3 33B is a 33.1B open weight model for coding agents and professional “cowork” tasks, listed under Apache 2.0.
What are the key points to validate first?
Announced on October 26, 2026, Nanbeige4.3 33B is a 33.1B open weight model for coding agents and professional “cowork” tasks, listed under Apache 2.0. Nanbeige also framed the 33B release as a teacher for smaller models, beginning with a planned Nanbeige4.3 3B.