DigestAI news desk

Cut through the AI noise.

Listen
Generative AI & Models3 min read

Nanbeige releases Nanbeige4.3-33B teacher model for lightweight agents

Nanbeige announced the Nanbeige4.3-33B model, a high‑capacity teacher designed to seed a pipeline that distills agentic abilities into smaller models such as a forthcoming 3B version. The 33B‑parameter model aims to set a new performance bar among open models in the ~30B parameter class, excelling at complex coding and professional cowork tasks.

1 source primary source

Key points

  • Nanbeige4.3-33B is a 33B‑parameter teacher model targeting lightweight agent distillation.
  • It outperforms Qwen3.8-27B, Muse‑Glimmer‑30B and Gemma4‑31B on most benchmarks, including 35 occupations in JobBench.
  • The model supports up to 524,288 tokens and can be served via Transformers, SGLang or vLLM.

The team evaluated Nanbeige4.3-33B against frontier models Qwen3.8-27B, Muse‑Glimmer‑30B and Gemma4‑31B using a suite of benchmarks that include Terminal‑Bench, DeepSWE, SWE‑Atlas‑QnA, NL2Repo, Automation Bench v1.0.6 (600 public tasks), JobBench (35 occupations), GDPval, Claw‑Eval General (157 tasks), Toolathlon‑Verified, YC‑bench (10 recent turns) and others. The model led on most tests and showed strong multi‑turn tool‑use performance when the “preservethinking” flag is enabled. It supports a context window of up to 524,288 tokens and can be run with Transformers, SGLang or vLLM.

Full story from huggingface.co · via Reddit AI communities primary sourceOpen source ↗

Nanbeige/Nanbeige4.3-33B · Hugging Face

huggingface.co · 25 October 2026

Loading the full article…

This text was published by huggingface.co. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Coverage and discussion

1source
Topics · follow one to build your own front page
NanbeigeHugging FaceNanbeige4.3-33BQwen3.8-27BMuse-Glimmer-30BGemma4-31BNanbeige Team

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Generative AI & Models

All →

Related stories

hot OpenAI's GPT-6 Astra: Changing Work and Research new Nvidia may charge $2,000 per GPU per year for AI software, report says hot Moonshot AI releases Kimi K3.1 Pro, a 2.8T open-weight model closing gap with proprietary leaders new Teradyne Robotics partners with Sereact to scale physical AI in retail hot PHP discloses critical PHP2Shell vulnerability affecting 70% of websites, discovered using GPT-5.6 Sol new NVIDIA says robotaxi leaders use its stack for $400B market hot OpenAI keeps frontier models offline after September DNS escape and red-teaming finds new OpenAI announces GPT-6.1 Sol with claimed 86.6% to 97.9% COBOL migration accuracy