Model reference · open weights
Kimi-Dev is an open-weight language model from moonshotai. Kimi-Dev-72B (BF16) weighs 145 GB; the smallest configuration that runs it is 4× L40S 48 GB.
Summary of the moonshotai/Kimi-Dev-72B model card, 2026-10-01
What it is
| Released by | moonshotai |
|---|---|
| Released | 2025-06-16 |
| Parameters | 72.7B |
| VRAM | 145 GB for the weights |
What it runs on
| Card | Requests at once | Context max | Memory | |
|---|---|---|---|---|
| 8K each | 32K each | |||
| RTX 3060 12 GB … H200 141 GB 11 smaller cards | — | — | — | |
| B200 180 GB | 8 | 2 | 69K | 176 GB |
| 4× L40S 48 GB tensor parallel | 9 | 2 | 79K | 44.0 GB a card |
| 2× RTX PRO 6000 Blackwell 96 GB tensor parallel | 11 | 2 | 90K | 93.8 GB a card |
| 2× H200 141 GB tensor parallel | 44 | 11 | all 128K | 138 GB a card |
| 4× H100 80 GB tensor parallel | 53 | 13 | all 128K | 78.1 GB a card |
| 4× A100 80 GB tensor parallel | 60 | 15 | all 128K | 78.2 GB a card |
| 2× B200 180 GB tensor parallel | 71 | 17 | all 128K | 176 GB a card |
| Requests at once | 8K tokens each | 32K tokens each |
|---|---|---|
| 1 | 149 GB | 157 GB |
| 5 | 160 GB | 200 GB |
| 8 | 168 GB | 232 GB |
| 16 | 189 GB | 318 GB |
| 32 | 232 GB | 490 GB |
| 64 | 318 GB | 834 GB |
One card, with vLLM's small-card settings.
From the model card
Introducing Kimi-Dev: A Strong and Open-source Coding LLM for Issue Resolution
We introduce Kimi-Dev-72B, our new open-source coding LLM for software engineering tasks. Kimi-Dev-72B achieves a new state-of-the-art on SWE-bench Verified among open-source models.
Kimi-Dev-72B achieves 60.4% performance on SWE-bench Verified. It surpasses the runner-up, setting a new state-of-the-art result among open-source models.
Kimi-Dev-72B is optimized via large-scale reinforcement learning. It autonomously patches real repositories in Docker and gains rewards only when the entire test suite passes. This ensures correct and robust solutions, aligning with real-world development standards.
Kimi-Dev-72B is available for download and deployment on Hugging Face and GitHub. We welcome developers and researchers to explore its capabilities and contribute to development.
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "moonshotai/Kimi-Dev-72B"
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
@misc{kimi_dev_72b_2025,
title = {Introducing Kimi-Dev: A Strong and Open-source Coding LLM for Issue Resolution},
author = {{Kimi-Dev Team}},
year = {2025},
month = {June},
url = {\url{https://www.moonshot.cn/Kimi-Dev}}
}
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.