AI Briefing
KOSign in

Nanbeige releases Nanbeige4.3-33B under Apache 2.0 with 512K context

·2026.10.26 01:32

Key point

The 33B model serves as a teacher for upcoming smaller agents and outperforms Qwen3.8-27B on Terminal-Bench 2.1 but trails on SWE-Bench Pro.

Details

  • Nanbeige released Nanbeige4.3-33B under the Apache 2.0 license as a teacher model for planned smaller agents, including an upcoming 3B model.
  • The model features a 512K context window and reports a score of 80.9 on Terminal-Bench 2.1, compared to 73.0 for Qwen3.8-27B using the same harness and inference settings.
  • Performance on SWE-Bench Pro is lower at 60.3 versus 61.7 for Qwen3.8-27B, indicating the model does not lead across all benchmarks.
  • Evaluations were conducted by the releasing team with preserve_thinking=true to retain earlier reasoning, a setting recommended for multi-turn tool use but not general chat.
  • Independent replication of the results has not been performed, and the team suggests comparing task success and total generated tokens with fixed settings for more practical deployment insights.
llmopen-sourcebenchmarkapache-2.0nanbeigeagent
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

Hacker NewsGVS5H: Zero-shot self-orchestration boosts Qwen3.8-27B to 92.4% on LiveCodeBenchPyTorchKRHugging Face Provides Unified API for Local and Cloud LLMs with AnyLanguageModel 1.0RedditCarbon-X: 14.8B DNA-Language Model Achieves 66.7% Benchmark Accuracy at 1/1,800th of Frontier CostRedditKapitsa Achieves 128 tok/s on MacBook Pro M4 Pro for Recursive's Flint 40B A2.5B Q4PyTorchKRPyMuPDF 2.0 Integrates Three PDF Preprocessing Packages and Applies GNN-Based Layout Analysis by DefaultGeekNewsKrafton Releases Open-Source Preview of Its In-House 30B-Class LLM 'Raon-1 mini'