Nanbeige4.2-3B
#9 in Small & Edge ModelsNanbeige · v4.2 · 3b · 3× · last seen Jul 24, 2026
42
Momentum
Nanbeige4.2-3B is a compact, agent-capable Open Source language model from Nanbeige (BOSS Zhipin) with 3 billion Non-Embedding parameters, built on Nanbeige4.2-3B-Base. It uses a Looped-Transformer architecture in which Transformer layers are traversed again after a complete pass to increase effective model capacity without additional parameters. According to the manufacturer, it outperforms larger models such as Qwen3.5-9B and Gemma4-12B on Tool-Use, Office-Agent, and Code-Agent Benchmarks. The model is distributed under the Apache-2.0 license and supports a context of up to 262,144 Token.
Momentum trend
26.04.25.07.
Features
| Key Benchmark (%) | SWE-Bench Verified: 63.6% (vs. Qwen3.5-9B 53.1%, Gemma4-12B 44.2%) |
| Context Window (Tokens) | 262,144 tokens (256K) |
| License | Apache 2.0 |
| Multimodality | Not multimodal – text-only language model (Text Generation) |
| Platform | Hugging Face; deployable via Transformers, vLLM, SGLang, llama.cpp, Ollama |
| Release Date | July 21, 2026 (announcement); model card updated late July 2026 |