Nanbeige Releases Open Weight Nanbeige4.3 33B Model Targeting Small Agent Coding and Tool Use
🚀Excited to announce Nanbeige4.3-33B! To push the performance ceiling of small agents, our strategy in Nanbeige4.3 starts with building a powerful teacher first. Nanbeige4.3-33B delivers SOTA agentic performance among open models at ~30B parameters: * Strong results on agentic coding and cowork benchmarks. * When integrated into mainstream agent harnesses, it tackles real-world tasks with cost-efficient inference. * It will serve as a teacher for upcoming smaller Nanbeige models, starting with Nanbeige4.3-3B. t.co/tSToQgZf2O t.co/D228euvybz #LLM #Agent
Context
Training a high performing 33 billion parameter model specifically to act as both a deployable coding agent and a distillation teacher pushes strong autonomous tool use down into weights small enough for single workstation GPUs.
The 30 billion parameter class has become a sweet spot for local and self hosted AI agents because quantized weights fit comfortably inside a single 24GB or 32GB consumer GPU while retaining enough reasoning capacity for multi step tool calls.
Deep dive
AI research lab Nanbeige announced and released Nanbeige4.3 33B on Sunday, a 33 billion parameter open weight language model built for agentic coding and workplace automation. According to the lab's announcement, its strategy for the Nanbeige 4.3 family started with training a capable medium sized teacher model first before distilling its capabilities into even smaller edge weights.
In self reported benchmark comparisons shared by Nanbeige, Nanbeige4.3 33B posts leading scores among open models in the 30 billion parameter tier across agentic software engineering, terminal use, and collaborative office benchmarks, outperforming similarly sized Qwen and Gemma releases on several multi turn agent evaluations. The team designed the model to plug directly into standard open source agent harnesses so developers can run multi step coding and file manipulation workflows at low inference cost.
Beyond serving as a standalone model for local and cloud deployments, Nanbeige stated that Nanbeige4.3 33B will act as the teacher model for a series of smaller student models, beginning with an upcoming 3 billion parameter release named Nanbeige4.3 3B aimed at on device agent tasks.
What to expect next
Independent evaluators and local LLM users will test Nanbeige4.3 33B inside harnesses such as Claude Code routers, OpenCode, and Cline to verify whether its real world coding reliability matches its self reported benchmark numbers, while watching for the upcoming 3B student release. If the 33 billion parameter teacher and its upcoming 3 billion parameter student hold up in community testing, local developers will gain a strong open weight option for running privacy sensitive coding agents on single GPU workstations and laptops.