The Heretic Organization's abliteration has both a halved refusal rate (4/100 vs. 8/100) and a lower KD divergence (0.0001 vs. 0.0003), therefore it is reccomended to use their version instead of this one.

Abliteration parameters

Parameter	Value
direction_index	per layer
attn.o_proj.max_weight	0.90
attn.o_proj.max_weight_position	27.04
attn.o_proj.min_weight	0.82
attn.o_proj.min_weight_distance	3.87
mlp.down_proj.max_weight	1.37
mlp.down_proj.max_weight_position	22.72
mlp.down_proj.min_weight	1.32
mlp.down_proj.min_weight_distance	14.90

Performance

Metric	This model	Original model (Nanbeige/Nanbeige4.1-3B)
KL divergence	0.0003	0 (by definition)
Refusals	8/100	100/100

Nanbeige Logo

Introduction

Nanbeige4.1-3B is built upon Nanbeige4-3B-Base and represents an enhanced iteration of our previous reasoning model, Nanbeige4-3B-Thinking-2511, achieved through further post-training optimization with supervised fine-tuning (SFT) and reinforcement learning (RL). As a highly competitive open-source model at a small parameter scale, Nanbeige4.1-3B illustrates that compact models can simultaneously achieve robust reasoning, preference alignment, and effective agentic behaviors.

Specifically, Nanbeige4.1-3B exhibits the following key strengths:

Strong Reasoning: Nanbeige4.1-3B is capable of solving complex, multi-step problems through sustained and coherent reasoning within a single forward pass, and reliably produces correct final answers on challenging tasks such as LiveCodeBench-Pro, IMO-Answer-Bench, and AIME 2026 I.
Robust Preference Alignment: Nanbeige4.1-3B achieves solid alignment performance, outperforming not only same-scale models such as Qwen3-4B-2507 and Nanbeige4-3B-2511, but also substantially larger models including Qwen3-30B-A3B and Qwen3-32B on Arena-Hard-v2 and Multi-Challenge.
Agentic Capability: Nanbeige4.1-3B is the first general small model to natively support deep-search tasks and reliably sustain complex problem solving involving more than 500 rounds of tool invocations. It fills a long-standing gap in the small-model ecosystem where models are typically optimized for either general reasoning or agentic scenarios, but rarely excel at both.

Technical Report: Coming Soon

Performances

We evaluate Nanbeige4.1-3B across a broad and diverse set of benchmarks covering general reasoning, and deep-search capabilities.

General Reasoning Tasks

On general reasoning tasks including code, math, science, alignment, and tool-use benchmarks, Nanbeige4.1-3B not only significantly outperforms same-scale models such as Qwen3-4B, but also demonstrates overall superior performance compared to larger models including Qwen3-30B-A3B-2507 and Qwen3-32B.

Benchmark	Qwen3-4B-2507	Qwen3-8B	Qwen3-14B	Qwen3-32B	Qwen3-30B-A3B-2507	Nanbeige4-3B-2511	Nanbeige4.1-3B
Code
Live-Code-Bench-V6	57.4	49.4	55.9	55.7	66.0	46.0	76.9
Live-Code-Bench-Pro-Easy	40.2	41.2	33.0	42.3	60.8	40.2	81.4
Live-Code-Bench-Pro-Mediium	5.3	3.5	1.8	3.5	3.5	5.3	28.1
Math
AIME 2026 I	81.46	70.42	76.46	75.83	87.30	84.1	87.40
HMMT Nov	68.33	48.33	56.67	57.08	71.25	66.67	77.92
IMO-Answer-Bench	48.00	36.56	41.81	43.94	54.34	38.25	53.38
Science
GPQA	65.8	62.0	63.38	68.4	73.4	82.2	83.8
HLE (Text-only)	6.72	5.28	7.00	9.31	11.77	10.98	12.60
Alignment
Arena-Hard-v2	34.9	26.3	36.9	56.0	60.2	60.0	73.2
Multi-Challenge	41.14	36.30	36.97	38.72	49.40	41.20	52.21
Tool Use
BFCL-V4	44.87	42.20	45.14	47.90	48.6	53.8	56.50
Tau2-Bench	45.9	42.06	44.96	45.26	47.70	41.77	48.57

Deep Search Tasks

As a general small model, Nanbeige4.1-3B achieves deep-search performance comparable to specialized agents under 10B parameters. In contrast to existing small general models, which typically exhibit little to no deep-search capability, Nanbeige4.1-3B represents a substantial qualitative improvement over prior small general models.

Deep Search and Agent Benchmarks

Model	xBench-DeepSearch-2505	xBench-DeepSearch-2510	Browse-Comp	Browse-Comp-ZH	GAIA (Text-only)	HLE	SEAL-0
Search-Specialized Small Agents
MiroThinker-v1.0-8B	61	–	31.1	40.2	66.4	21.5	40.4
AgentCPM-Explore-4B	70	–	25.0	29.0	63.9	19.1	40.0
Large Foundation Models (with Tools)
GLM-4.6-357B	70	–	45.1	49.5	71.9	30.4	–
Minimax-M2-230B	72	–	44.0	48.5	75.7	31.8	–
DeepSeek-V3.2-671B	71	–	67.6	65.0	63.5	40.8	38.5
Small Foundation Models (with Tools)
Qwen3-4B-2507	34	5	1.57	7.92	28.33	11.13	15.74
Qwen3-8B	31	2	0.79	5.15	19.53	10.24	6.34
Qwen3-14B	34	9	2.36	7.11	30.23	10.17	12.64
Qwen3-32B	39	8	3.15	7.34	30.17	9.26	8.15
Qwen3-30B-A3B-2507	25	10	1.57	4.12	31.63	14.81	9.24
Ours (with Tools)
Nanbeige4-3B-2511	33	11	0.79	3.09	19.42	13.89	12.61
Nanbeige4.1-3B	75	39	19.12	31.83	69.90	22.29	41.44

Quickstart

For inference hyperparameters, we recommend the following settings:

Temperature: 0.6
Top-p: 0.95
Repeat penalty: 1.0
Max New Tokens: 131072

For the chat scenario:

from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(
  'megabytes/Nanbeige4.1-3B-heretic',
  use_fast=False,
  trust_remote_code=True
)
model = AutoModelForCausalLM.from_pretrained(
  'megabytes/Nanbeige4.1-3B-heretic',
  torch_dtype='auto',
  device_map='auto',
  trust_remote_code=True
)
messages = [
  {'role': 'user', 'content': 'Which number is bigger, 9.11 or 9.8?'}
]
prompt = tokenizer.apply_chat_template(
  messages,
  add_generation_prompt=True,
  tokenize=False
)
input_ids = tokenizer(prompt, add_special_tokens=False, return_tensors='pt').input_ids
output_ids = model.generate(input_ids.to('cuda'), eos_token_id=166101)
resp = tokenizer.decode(output_ids[0][len(input_ids[0]):], skip_special_tokens=True)
print(resp)

For the tool use scenario:

from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(
  'megabytes/Nanbeige4.1-3B-heretic',
  use_fast=False,
  trust_remote_code=True
)
model = AutoModelForCausalLM.from_pretrained(
  'megabytes/Nanbeige4.1-3B-heretic',
  torch_dtype='auto',
  device_map='auto',
  trust_remote_code=True
)
messages = [
    {'role': 'user',  'content': 'Help me check the weather in Beijing now'}
]
tools = [{'type': 'function',
  'function': {'name': 'SearchWeather',
   'description': 'Find out the current weather in a place on a certain day.',
   'parameters': {'type': 'dict',
    'properties': {'location': {'type': 'string',
      'description': 'A city in China.'},
    'required': ['location']}}}}]
prompt = tokenizer.apply_chat_template(
  messages,
  tools,
  add_generation_prompt=True,
  tokenize=False
)
input_ids = tokenizer(prompt, add_special_tokens=False, return_tensors='pt').input_ids
output_ids = model.generate(input_ids.to('cuda'), eos_token_id=166101)
resp = tokenizer.decode(output_ids[0][len(input_ids[0]):], skip_special_tokens=True)
print(resp)

For the deep-search scenario:

Inference Framework: miroflow-framework!
Switch tokenizer configuration to tokenizer_config_search.json
Tools Configuration:

Server	Description	Tools Provided
tool-python	Execution environment and file management (E2B sandbox)	create_sandbox, run_command, run_python_code, upload_file_from_local_to_sandbox, download_file_from_sandbox_to_local, download_file_from_internet_to_sandbox
search_and_scrape_webpage	Google search via Serper API	google_search
jina_scrape_llm_summary	Web scraping with LLM-based information extraction with Jina	scrape_and_extract_info

Summary model: Qwen3-14B-thinking
Temperature: 1.0
Note, access to HuggingFace has been explicitly disabled in these tools.

Limitations

While we place great emphasis on the safety of the model during the training process, striving to ensure that its outputs align with ethical and legal requirements, it may not completely avoid generating unexpected outputs due to the model's size and probabilistic nature. These outputs may include harmful content such as bias or discrimination. Please don't propagate such content. We do not assume any responsibility for the consequences resulting from the dissemination of inappropriate information.

Contact

If you have any questions, please raise an issue or contact us at nanbeige@kanzhun.com.