Model library / Nanbeige
Nanbeige4.2-3B is a compact agentic language model designed for reasoning, tool use and personal-assistant tasks.
Parameters in total
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
Nanbeige4.2-3B is a 3 billion-parameter agentic model built on the Nanbeige4.2-3B-Base foundation. It uses a Looped Transformer architecture to reuse layers and increase capacity without adding parameters. The publisher positions it for strong agentic behaviour, broad reasoning and alignment capabilities, targeting multi-step tool-use, office workflows and code-assistant scenarios.
In practice the model contains 848 million trainable parameters and supports a context window of up to 262 144 tokens. Weights are stored in bfloat16 format and the model is released under the Apache-2.0 licence. It has no expert layers, making it straightforward to run on standard GPU hardware for local deployment.
Specification
| Repository | Nanbeige/Nanbeige4.2-3B-DSpark |
| Publisher | Nanbeige |
| Published | 31 August 2026 |
| Architecture | Qwen3DSparkModel |
| Model type | qwen3 |
| Parameters | 847,946,241 |
| Layers | 5 |
| Hidden size | 3,072 |
| Attention heads | 48, 8 key/value heads |
| Context length | 262,144 tokens |
| Vocabulary | 166,144 tokens |
| Weight format | bfloat16 |
| Licence | apache-2.0 |
| Base model | Nanbeige/Nanbeige4.2-3B-Base |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 1.6 GiB | Full precision as released |
| fp8 | 0.8 GiB | 8-bit, near-lossless on most models |
| int4 | 0.4 GiB | 4-bit, smallest footprint |
The cache costs 20 KiB per token at 16-bit, so the full 262,144-token context of one request needs about 5.0 GiB. Concurrency multiplies that number, not the weights.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 1 | 17.5 GiB | the full 262,144 tokens |
| RTX 5090 32GB | 1 | 24.7 GiB | the full 262,144 tokens |
| L40S 48GB | 1 | 39.1 GiB | the full 262,144 tokens |
| A100 80GB | 1 | 67.9 GiB | the full 262,144 tokens |
| H100 80GB | 1 | 67.9 GiB | the full 262,144 tokens |
| RTX PRO 6000 96GB | 1 | 82.3 GiB | the full 262,144 tokens |
| H200 141GB | 1 | 122.8 GiB | the full 262,144 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve Nanbeige/Nanbeige4.2-3B-DSpark \ --max-model-len 262144 \ --gpu-memory-utilization 0.90
Questions
The weights take about 2 GiB at the released precision. That fits on one RTX 4090 24GB, which leaves roughly 18 GiB for the key-value cache.
The repository declares apache-2.0. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 262,144 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email [email protected] with the model name and roughly what volume you expect.
Also in the library
Nanbeige4.2-3B is a 4.2 billion-parameter causal language model aimed at compact agentic and reasoning tasks.
K2-Horizon-0.9B is a 1.1 billion-parameter dense decoder-only model with a 128 k token context window.
S1-mini is a 0.6 B-parameter text normaliser that cleans raw English ASR output into properly punctuated, capitalised written tex…
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.