AI Briefing
KOSign in

Nanbeige4.3-33B ranks first on 17 of 20 benchmarks

·2026.10.28 11:30

Key point

Kanzhun released the 33B dense model Nanbeige4.3-33B under the Apache 2.0 license.

1 / 4

Details

  • The Kanzhun Nanbeige team released the Nanbeige4.3-33B weights on Hugging Face under the Apache 2.0 license on October 24, 2026
  • A 33.1B parameter dense model designed as a teacher model for distilling agent capabilities into a 3B small model to be released in the future
  • Recorded superiority on 17 of 20 benchmarks compared to Qwen3.8-27B, Muse-Glimmer-30B, and Gemma4-31B
  • Achieved first place in agent coding and reasoning metrics, including Terminal-Bench 2.1 80.9 and AIME 2026 95.2
  • Supports 512K context, text-only input, and English and Chinese
  • History of changing the initial public 1M context setting to 512K the next day
  • Technical report and training data not disclosed, Korean evaluation results not provided
llmopen-sourcebenchmarkdistillationnanbeigeagentkangzun
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditNanbeige releases Nanbeige4.3-33B under Apache 2.0 with 512K contextRedditinclusionAI releases Ling-3.1-flash-1019 with 560B parameters and 1M context windowHacker NewsGVS5H: Zero-shot self-orchestration boosts Qwen3.8-27B to 92.4% on LiveCodeBenchPyTorchKRBespoke Labs Releases Decision Model NimblePyTorchKRStepFun Step Code adds built-in sub-agents and web deployment featuresHacker NewsOpenAI releases GPT-6.1 Terra with 50% lower API costs and improved computer use