Home › Language Models › Kimi-Dev

Kimi-Dev - AI Language Models Tool

Overview

Kimi-Dev (Kimi-Dev-72B) is an open-source coding large language model from Moonshot AI focused on automated software engineering tasks — especially automated bug repair and unit-test generation. The model is released with model weights and code (MIT license) and was trained and fine-tuned from a Qwen-2.5-72B base through a two-stage pipeline: extensive mid-training on real GitHub issues/PR data and a large-scale reinforcement-learning phase that applies candidate patches inside Docker and rewards only when the full test suite passes. ([huggingface.co](https://huggingface.co/moonshotai/Kimi-Dev-72B?utm_source=openai)) Kimi-Dev uses a duo design (BugFixer + TestWriter) that first localizes files to edit and then produces edits or unit tests; at test time it can self-play between those roles and scale by generating many patch/test candidates. The project provides runnable tooling (examples, vLLM serving instructions) for agentless rollouts and encourages community contributions via GitHub and Hugging Face. Reported headline metrics include a 60.4% resolve rate on SWE-bench Verified; community feedback praises its debugging accuracy while noting inference speed, quantization trade-offs, and real-world variability. ([github.com](https://github.com/MoonshotAI/Kimi-Dev?utm_source=openai))

GitHub Statistics

  • Stars: 1,148
  • Forks: 147
  • Contributors: 3
  • License: NOASSERTION
  • Primary Language: Python
  • Last Updated: 2025-09-30T02:15:54Z

Key Features

  • Autonomous repository patching in Docker; rewards only when full test suites pass.
  • Dual BugFixer + TestWriter workflow for paired fixes and unit-test generation.
  • Large-scale RL fine-tuning with outcome-only rewards to prioritize correctness.
  • Very long context support (131,072 tokens) for multi-file or large-repo reasoning.
  • Open-source weights, vLLM-ready serving instructions, and example rollout scripts.

Example Usage

Example (python):

from transformers import AutoModelForCausalLM, AutoTokenizer

# Example quick-start (adapted from the project's model card). See HF model page for details.
model_name = "moonshotai/Kimi-Dev-72B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")

prompt = "# Fix the failing test: add the missing edge-case handling in function foo\n"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

# Note: for production or multi-request serving, the repo provides vLLM serving examples and rollout scripts. ([huggingface.co](https://huggingface.co/moonshotai/Kimi-Dev-72B?utm_source=openai))

Benchmarks

SWE-bench Verified resolve rate: 60.4% (Source: https://huggingface.co/moonshotai/Kimi-Dev-72B)

Model family / nominal size: Kimi-Dev-72B (72B parameters, HF lists ~73B) (Source: https://huggingface.co/moonshotai/Kimi-Dev-72B)

Context window (max tokens): 131,072 tokens (131K) (Source: https://github.com/MoonshotAI/Kimi-Dev)

Mid-training data volume: ~150B training tokens (mid-training on GitHub issues/PRs) (Source: https://moonshotai.github.io/Kimi-Dev/)

Last Refreshed: 2026-02-24

Key Information

  • Category: Language Models
  • Type: AI Language Models Tool
Visit Official Website

Related Tools

Qwen2.5-7B

Qwen2.5-7B is a large language model designed for text generation, featuring improvements in coding, mathematics, instruction following, long text generation, and multilingual support. It supports context lengths up to 128K tokens and is intended for sophisticated NLP tasks.
DeepSeek-V3

Open‑weight flagship LLM (base and chat) with strong reasoning and tool‑use; runs with SGLang/LMDeploy/vLLM and permissive model license.
Llama 3

Llama 3 is an open access large language model (LLM) released by Meta, available in various configurations (8B and 70B parameters) with capabilities for fine-tuning and integrations into platforms like Hugging Face, Google Cloud, and Amazon SageMaker.
UNfilteredAI-1B

A large-scale text generation model designed for creative and unconstrained content production without traditional filtering.
Shuttle-3

Shuttle-3 is a state-of-the-art language model designed for high-quality text generation, particularly suited for complex chat, multilingual communication, and reasoning tasks. It is fine-tuned from the Qwen-2.5-72b-Instruct model and designed to emulate high-quality prose similar to Claude 3 models.
WizardLM

WizardLM is a state-of-the-art large language model designed for complex chat, multilingual tasks, reasoning, and agent functionalities. It features an AI-powered pipeline (Auto Evol-Instruct) that optimizes instruction datasets for improved performance across various domains and leverages Arena Learning for an expanded learning pool of challenging instruction data.