Nanbeige releases open 33B agentic model Nanbeige4.3-33B under Apache 2.0
Article timeline
The articles in this story. Times are ET.
ModelScopeOfficialNanbeige4.3-33B GGUF offers quantized local inference for coding and reasoningAIModelScope introduces Nanbeige4.3-33B GGUF, a 33B open-weight dense model with quantized weights for local inference. The base model supports up to 512K tokens, with usable local context depending on runtime settings and memory, and the post says it targets repository refactoring, terminal tasks, and multi-step agent workflows. Q4_K_M, Q5_K_M, Q6_K, and Q8_0 options are available under Apache 2.0, with llama.cpp and Ollama setup examples.
Thomas WolfXPickNanbeige4.3-33B open-weights model posts strong agentic benchmark results against Qwen3.8AIThomas Wolf says BOSS Zhipin's Nanbeige research team released Nanbeige4.3-33B as an open-weights model. The quoted post says it leads open models near 30B parameters on agentic benchmarks and will serve as a teacher for smaller Nanbeige models, starting with Nanbeige4.3-3B. The attached table shows it scoring higher than Qwen3.8 27B on most agentic coding and cowork tests but not on SWE-Bench Pro, GDPval, or GPQA Diamond.
- PandailyNewsBOSS Zhipin's Nanbeige team open-sources Nanbeige4.3-33B, a 512K-context agentic coding model
AIKanzhun's Nanbeige team has open-sourced Nanbeige4.3-33B under Apache 2.0, a model built for agentic coding and office tasks. The team says it leads most of the benchmarks it compares against roughly 30B open models, though these results are self-reported and not independently verified. The model supports up to 524,288 tokens of context and runs with Transformers, SGLang or vLLM.
- ModelScopeOfficialNanbeige4.3-33B releases with 512K context and agentic benchmark results
AIModelScope announces Nanbeige4.3-33B, a 33B dense model with 512K context, and says it leads most benchmarks against Qwen3.8-27B, Muse-Glimmer-30B, and Gemma4-31B. It reports 80.9 on Terminal-Bench 2.1 and 65.5 on DeepSWE, versus 73.0 and 42.5 for Qwen3.8-27B. The model is licensed under Apache 2.0, supports Transformers, SGLang, and vLLM, and is planned as a teacher model for distillation into smaller models.
- MarkTechPostNewsNanbeige releases Nanbeige4.3-33B, an Apache 2.0 agentic coding model
AINanbeige LLM Lab released Nanbeige4.3-33B on Hugging Face under the Apache 2.0 license. The 33B dense, text-only model supports a 524,288-token context window and is built for agentic coding and office-style cowork tasks. In Nanbeige's own comparison, it leads 17 of 20 benchmark rows against three open models near 30B, scoring 80.9 on Terminal-Bench 2.1. Several competitor figures in that table differ from the originating labs' model cards.
🚨 AI News | TestingCatalogXNanbeige releases Nanbeige4.3-33B, a ~30B open model for agentic tasksAINanbeige releases Nanbeige4.3-33B, an open model in the roughly 30B parameter class that it says leads most benchmarks in agentic coding and cowork tasks. The company says the model outperforms Qwen3.8-27B, Muse-Glimmer-30B, and Gemma4-31B on most benchmarks. Nanbeige plans to use it as a teacher for a smaller Nanbeige4.3-3B model.
