Unsloth
FreemiumRun and train AI models on your own computer.
Published 27 September 2026 · Last updated 1 October 2026
Scores
Popularity4/5
One of the most-starred projects in the fine-tuning space and a common default in tutorials, with widely used quantized model releases on Hugging Face.
Learning Curve3/5
The prebuilt Colab and Kaggle notebooks make a first fine-tune approachable, but producing a genuinely good model still requires dataset preparation, hyperparameter tuning, and evaluation skill.
Flexibility4/5
Covers LoRA, QLoRA, full fine-tuning, pretraining, and RL methods across many model families, though it is purpose-built for fine-tuning rather than a general training framework.
Performance5/5
Hand-written fused kernels deliver roughly 70% memory reduction and about 2x faster training with no accuracy loss, best-in-class for efficient fine-tuning.
Portability5/5
Apache 2.0 licensed, self-hostable on any compatible GPU, and exports adapters to GGUF, Ollama, and vLLM, so trained models run anywhere with no lock-in.
About Unsloth
Unsloth is an open-source tool for running and training AI models on your own hardware. It comes in three forms that share one engine: Unsloth Desktop, a native app for Windows, macOS, and Linux; Unsloth Studio, a local web UI started from the command line or a Docker image; and Unsloth Core, the Python library it began as. All three work with open-weight models from Hugging Face, including Llama, Qwen, Gemma, DeepSeek, and Mistral, along with diffusion, embedding, and audio models.
For running models it works as a local chat app and inference server. It loads GGUF and MLX builds, running GGUF files on llama.cpp, and exposes them through an OpenAI-compatible API so other tools can use a local model by changing a base URL. A single command connects coding agents such as Claude Code, Codex, and OpenCode to a local model, and the app adds web search, document retrieval, tool calling, and access from other devices on the network.
Training is what Unsloth is best known for. Its hand-written fused kernels rewrite the compute-heavy parts of the training stack, using roughly 70% less memory and running about twice as fast as a standard setup, which makes fine-tuning practical on a single consumer or cloud GPU. It supports LoRA, QLoRA, full fine-tuning, continued pretraining, and reinforcement-learning methods such as GRPO and DPO. Data Recipes build datasets from PDFs, CSVs, and documents without code, and trained models export to GGUF and safetensors for serving in Ollama, llama.cpp, or vLLM.
The core package is Apache 2.0 licensed and the Studio interface is AGPL-3.0; both are free to use. Unsloth AI sells Pro and Enterprise plans for faster training at larger scale and for support. Compared with LM Studio, which only runs models, Unsloth also trains them; compared with Ollama, it leads with a graphical app instead of a command line.
Key Features
- Unsloth Desktop for Windows, macOS, and Linux, plus the Studio web UI and a Python library
- Runs GGUF and MLX models locally and serves them through an OpenAI-compatible API
- One command connects Claude Code, Codex, OpenCode, and other agents to a local model
- Fused training kernels: roughly 70% less VRAM and about 2x faster fine-tuning
- LoRA, QLoRA, full fine-tuning, continued pretraining, and RL methods (GRPO, DPO)
- Data Recipes build training datasets from PDFs, CSVs, and DOCX files
- Export to GGUF and safetensors for Ollama, llama.cpp, and vLLM
- Built-in web search, document retrieval, and tool calling for local models
Pros
- Makes fine-tuning viable on a single consumer or cloud GPU
- One app covers running, chatting with, training, and exporting a model
- Open source throughout, unlike LM Studio's closed desktop app
- Ready-made Colab and Kaggle notebooks lower the barrier to a first fine-tune
- Runs on NVIDIA, AMD, and Intel GPUs, Apple Silicon, and CPU-only machines
Cons
- Reviews find its chat interface less polished than apps built only for running models
- Fine-tuning still requires understanding of datasets, hyperparameters, and evaluation
- The Studio interface is AGPL-3.0, a different licence from the Apache 2.0 core
- Pricing for the Pro and Enterprise plans is not clearly published
- Support for brand-new models can arrive faster than stable documentation
Unsloth Pricing
Freemium- · Unsloth Desktop, Unsloth Studio, and the core library, free to use
- · Local inference, fine-tuning, dataset building, and export
- · Faster training and lower memory use than the open-source version
- · Contact Unsloth for pricing
- · Multi-node training and dedicated support
- · Custom quote from sales
Tech Stacks with Unsloth
Local LLM Dev Stack
DeveloperA coding agent that works against a model on your own hardware. A local runtime such as Ollama or LM Studio serves an open-weight coding model such as Qwen or Mistral's Devstral, and an agent such as OpenCode or Cline does the editing. No code leaves the machine and there is no per-token bill.
Tools Related to Unsloth
Works well with Unsloth(11)
Unsloth offers optimized fine-tuning recipes and quantized releases for Qwen models.
Unsloth provides optimized, memory-efficient fine-tuning support for Llama models, one of its most common training targets.
Unsloth publishes GGUF quantizations of Xiaomi MiMo and can fine-tune it, making the model runnable on smaller hardware.
Unsloth publishes fine-tuning notebooks and Dynamic GGUF builds for the DeepSeek V3/R1 line and successors.
Unsloth ships Dynamic GGUFs and fine-tuning support for the GLM family, GLM-4.5 and newer.
Unsloth maintains Dynamic GGUF builds and fine-tuning guides across the Gemma line.
Built on (1)
Unsloth is a Python library for faster, lower-memory LLM fine-tuning.
Required by Unsloth(1)
Unsloth is built on PyTorch; its optimized training kernels and custom autograd run inside PyTorch, and there is no other backend.
Alternatives to Unsloth(2)
LlamaFactory covers more training methods and model families, driven from YAML configs or a no-code web UI, and scales to multi-GPU clusters; Unsloth is faster and lighter on memory on a single GPU and also runs models locally. LlamaFactory can use Unsloth's kernels as a backend, but a project picks one as its toolkit.
Unsloth runs local models and also fine-tunes them, builds datasets, and exports the result, all as open source; LM Studio is a closed-source app focused on finding, chatting with, and serving models. Pick Unsloth to train as well as run, LM Studio for the more polished everyday chat app.
Vendor
Unsloth AI
Website →Tags
Details
- Maintained
- Yes