Nanbeige/Nanbeige4.3-33B · Hugging Face

Source: https://huggingface.co/Nanbeige/Nanbeige4.3-33B

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#1-introduction1. Introduction

In Nanbeige4.3, we take a foundational step toward pushing the boundaries of lightweight agents. Our strategy centers on building a high-capacity teacher agent first, which will subsequently anchor the distillation of advanced agentic capabilities into smaller models.

To this end, we first introduce Nanbeige4.3-33B, designed as the flagship teacher model for the Nanbeige4.3 series. Delivering state-of-the-art agentic performance among open models in the ~30B parameter class, Nanbeige4.3-33B excels at complex coding and professional cowork tasks. It not only offers a powerful, cost-effective option for immediate real-world deployment, but also lays the necessary foundation to power our upcoming small model distillation pipeline (e.g., for Nanbeige4.3-3B).

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#2-model-performance2. Model Performance

We compare Nanbeige4.3-33B with recently released frontier models at around the 30B-parameter scale: Qwen3.8-27B, Muse-Glimmer-30B, and Gemma4-31B. All models are evaluated at their highest thinking effort. For each benchmark, we use the same inference hyperparameters and harness across models.

CategoryBenchmarkNanbeige4.3-33B1Qwen3.8-27BMuse-Glimmer-30BGemma4-31BAgentic CodingTerminal-Bench 2.1280.973.061.847.2Terminal-Bench 4.0323.86.31.60.0DeepSWE365.542.54.42.7SWE-Bench Pro360.361.736.951.2SWE-Atlas-QnA450.841.917.73.3NL2Repo553.048.327.433.8Scicode52.646.744.945.5SkillsBench61.156.439.728.9Agentic CoworkAutomation Bench654.238.315.29.7JobBench744.834.516.69.0GDPval884.886.475.871.4Claw-Evalpass^3978.375.257.344.2Toolathlon-Verified1070.468.547.227.8DeepPlanning1263.349.221.419.4EnterpriseOps-Gym1241.040.230.527.3YC-bench11**1,025,437**641,290513,978265,152Tau3-Bench1248.848.533.718.6ReasoningGPQA Diamond89.589.283.591.3AIME 202695.294.594.789.2AA-LCR84.082.083.372.01Nanbeige4.3-33B is evaluated withpreserve\_thinking=truein the chat template. 2Terminal-Bench 2.1 is evaluated using Terminus 2. 3Terminal-Bench 4 and DeepSWE are evaluated using mini-swe-agent v2.4.6. 4SWE-Atlas-QnA is evaluated using mini-swe-agent v2.4.6. Answers are evaluated against 1,304 task-specific rubrics using GPT-6 Sol as the judge, with medium reasoning effort. 5NL2Repo is evaluated using Claude Code. 6Automation Bench v1.0.6 includes 600 publicly released tasks. 7JobBench is evaluated with the Claude Code harness equipped with a search tool. 8GDPval is scored using revised rubrics. 9Claw-Eval is evaluated exclusively on the Claw-Eval General subset (comprising 157 single-turn user interaction tasks). 10Toolathlon-Verified sets no max_tokens limit for individual model requests. 11YC-bench retains only the most recent 10 turns of user dialogue as historical context before each model call. 12DeepPlanning, EnterpriseOps-Gym, Tau3-Bench are evaluated using their official repositories.

Nanbeige4.3-33B leads on most of the benchmarks, showing strong performance in both agentic coding and cowork tasks. Beyond its breadth across benchmarks, Nanbeige4.3-33B shows strong performance onagent tasks across a wide range of professions and industries. For example, on JobBench, it outperforms the baseline in most of the 35 occupations evaluated.

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#3-quickstart3. Quickstart

The tokenizer provides a configurable chat template for reasoning and tool-use scenarios:

• enable\_thinkingcontrols whether the model generates reasoning for the current response. It is enabled by default; set it toFalsefor non-thinking mode.
• preserve\_thinkingcontrols whether reasoning from previous assistant turns is retained in a multi-turn conversation. We recommendFalsefor general chat and question answering, andTruefor multi-turn tool use, office tasks, and code-agent workflows.
• Passingtoolsenables the tool-use template. We recommendtool\_call\_format="xml"for the best tool-calling performance;jsonis also supported for compatibility.

The model supports a context length of up to 524,288 tokens (512K).

We recommend adjusting the inference settings according to the target scenario:

ScenarioTemperatureTop_pMax New TokensAgentic and tool-use tasks1.00.9565,536Reasoning and chat tasks0.60.95262,144 You can run the model with Transformers, SGLang, or vLLM:

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#transformersTransformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = 'Nanbeige/Nanbeige4.3-33B'
tokenizer = AutoTokenizer.from_pretrained(
  model_id,
  use_fast=False,
  trust_remote_code=True
)
model = AutoModelForCausalLM.from_pretrained(
  model_id,
  torch_dtype='auto',
  device_map='auto',
  trust_remote_code=True
)

messages = [{'role': 'user', 'content': 'Which number is bigger, 9.11 or 9.8?'}]
inputs = tokenizer.apply_chat_template(
  messages,
  add_generation_prompt=True,
  return_tensors='pt'
).to(model.device)
output_ids = model.generate(inputs, max_new_tokens=1024)
response = tokenizer.decode(output_ids[0][inputs.shape[-1]:], skip_special_tokens=True)
print(response)

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#sglangSGLang

MODEL_PATH=/path/to/your/Nanbeige4.3-33B
python -m sglang.launch_server \
  --model-path ${MODEL_PATH} \
  --host 0.0.0.0 \
  --port 8000 \
  --tp-size 4 \
  --mem-fraction-static 0.8 \
  --reasoning-parser qwen3 \
  --tool-call-parser qwen3_coder

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#vllmvLLM

MODEL_PATH=/path/to/your/Nanbeige4.3-33B
vllm serve ${MODEL_PATH} \
  --host 0.0.0.0 \
  --port 8000 \
  --tensor-parallel-size 4 \
  --gpu-memory-utilization 0.8 \
  --enable-auto-tool-choice \
  --reasoning-parser qwen3 \
  --tool-call-parser qwen3_coder

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#4-contact4. Contact

If you have any questions, please open an issue in this repository or contact us at[email protected].

We also welcome collaborations to build domain-specific and industry-focused agents, including joint work on data and training.

https://huggingface.co/Nanbeige/Nanbeige4.3-33B#5-citation5. Citation

If you find our model useful or would like to use it in your own work, please cite this Hugging Face project:

@misc{nanbeige43chat2026,
  title        = {Nanbeige4.3-33B},
  author       = {Nanbeige Team},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Nanbeige/Nanbeige4.3-33B}}
}