We are thrilled to introduce Stable-DiffCoder, which is a strong code diffusion large language model. Built directly on the Seed-Coder architecture, data, and training pipeline, it introduces a block diffusion continual pretraining (CPT) stage with a tailored warmup and block-wise clipped noise schedule.
Under identical architecture and data settings, we systematically analyze and design an efficient diffusion training pipeline that is not only stable but also potentially lifts the model’s performance ceiling. With this recipe, Stable-DiffCoder demonstrates overall performance improvements compared to its autoregressive (AR) counterpart across a broad set of code benchmarks, while any-order modeling improves structured code handling for editing and reasoning, and diffusion-based corruption aids learning for low-resource programming languages.
Notably, with only CPT followed by supervised fine-tuning, Stable-DiffCoder further surpasses many strong ∼8B AR and diffusion-based code models. These results demonstrate that diffusion-based training can improve code modeling quality beyond what AR training alone can achieve, even under tightly controlled data and architecture constraints.
This repo contains the Stable-DiffCoder-8B-Instruct model, which has the following features:
Current (v5.3.0) transformers is available for inference:
pip install transformers~=5.3.0
Explanation of Inference Parameters
steps: Number of steps for diffusion generation
gen_length: Maximum length of the generated output
block_length: Length of the diffusion block, with a default value of 4
temperature: Temperature for generation, with a default value of 0.0
remasking: Remasking strategy, optional values are 'low_confidence' or 'random', default value is 'low_confidence' (for principle, refer to LLADA)
tokenizer: Tokenizer used for text encoding and decoding
shift: Whether to shift the output to the right by one position (similar to AutoRegressive/AR), default value is False
threshold: Threshold for decoding (range: 0-1.0), default value is None; a smaller value results in faster decoding speed (for principle, refer to Fast-DLLM)
eos_id: ID of the end-of-sequence token, default value is tokenizer.eos_token_id
Quickstart
Here is a simple example demonstrating how to load the model and generate code.
Stable-DiffCoder-8B-Instruct has been evaluated on a wide range of coding tasks, including code generation, code reasoning, code editing, achieving stronger performance than
a wide range of ∼8B ARs and DLLMs,
Compared with ∼8B AR models：
Model
HumanEval
MBPP
MHPP
BigCodeBench (Full)
BigCodeBench (Hard)
LiveCodeBench (v5)
CodeLlama-7B-Instruct
40.9
54.0
6.7
25.7
4.1
3.6
DeepSeek-Coder-6.7B-Instruct
74.4
74.9
20.0
43.8
15.5
9.6
CodeQwen1.5-7B-Chat
83.5
77.7
17.6
43.6
15.5
3.0
Yi-Coder-9B-Chat
82.3
82.0
26.7
49.0
17.6
17.5
Llama-3.1-8B-Instruct
68.3
70.1
17.1
40.5
13.5
11.5
OpenCoder-8B-Instruct
83.5
79.1
30.5
50.9
18.9
17.1
Qwen2.5-Coder-7B-Instruct
88.4
83.5
26.7
48.8
20.3
17.3
Qwen3-8B
84.8
77.0
32.8
51.7
23.0
23.5
Seed-Coder-8B-Instruct
84.8
85.2
36.2
53.3
26.4
24.7
Stable-DiffCoder-8B-Instruct
86.6
85.7
42.4
54.8
31.8
23.5
Compared with ∼8B DLLM models：
Model
HumanEval
HumanEval+
MBPP
MBPP+
BigCodeBench (Full)
LLaDA-8B-Instruct
49.4
-
41.0
-
16.5
Dream-7B-Instruct
63.4
-
68.3
-
10.6
LLaDA-MoE-7B-Instruct
61.6
-
70.0
-
20.4
Fast-dLLMv2
43.9
40.2
50.0
41.3
49.0
DiffuCoder-7B-Instruct
72.0
65.2
75.1
61.9
35.7
Dream-Coder-7B-Instruct
82.9
-
79.6
-
37.1
SDAR-8B-Chat
78.7
-
72.0
-
-
WeDLM-8B-Chat
80.5
73.8
70.5
-
-
Stable-DiffCoder-8B-Instruct
86.6
82.3
85.7
72.8
54.8
For detailed benchmark performance, please refer to our 📑 Technical Report.
License
This project is licensed under the MIT License. See the LICENSE file for details.
Citation
If you find our work helpful, feel free to give us a cite.
@misc{fan2026stablediffcoderpushingfrontiercode,
title={Stable-DiffCoder: Pushing the Frontier of Code Diffusion Large Language Model},
author={Chenghao Fan and Wen Heng and Bo Li and Sichen Liu and Yuxuan Song and Jing Su and Xiaoye Qu and Kai Shen and Wei Wei},
year={2026},
eprint={2601.15892},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2601.15892},
}
