unsloth
Unsloth is a local UI for training and running Kimi K3, Gemma 4, Qwen3.6, DeepSeek-V4, GLM and other models.
The Lens
By Erik Loyd, SaaS CEO and former COO/CFO of an AWS Premier Partner.
Updated Oct 2026
Unsloth is the free way to fine-tune open models on your own hardware: teaching an existing model like Qwen, Gemma or gpt-oss your data, about 2x faster on 70% less VRAM, per the project. It now comes three ways: Unsloth Desktop (a native app, in beta), Unsloth Studio (the same tool as a web UI) and Unsloth Core (the Python library).
Desktop installs like any app on Mac, Windows or Linux and runs models too: GGUF and MLX chat, image and video generation, an OpenAI-compatible API, and unsloth start claude to point Claude Code at a local model. Training works on NVIDIA, AMD, Intel and Mac. CPU-only machines get chat and dataset building, not training.
All of that is free. Unsloth also sells Pro and Enterprise builds for multi-GPU and multi-node speedups, contact sales only, on a pricing page that still calls multi-GPU 'coming soon'. Solo and small teams: free. Large training clusters: talk to sales.
The catch is licensing. The pip package says Apache 2.0, but Studio's UI, which Desktop wraps, is AGPL-3.0, and so are the mixture-of-experts kernels inside Core. Training your own models is unaffected. Shipping a product on top of it is not.
Free vs Self-Hosted vs Paid
open coreFree: Unsloth Desktop, Studio and Core are free: LoRA, QLoRA, full fine-tuning, pretraining and RL (GRPO, DPO) for 500+ text, vision, TTS and embedding models, 500K-context training, multi-GPU per the README, Data Recipes for building datasets from PDFs and CSVs, local inference (GGUF, MLX, diffusion image and video), an OpenAI-compatible API, web search, and export to GGUF and other formats. Free Colab notebooks cover most models up to 22B on T4 GPUs.
Self-hosted: Everything runs locally. Native Desktop installers for macOS, Windows and Linux (including ARM64), an install script, Homebrew, winget, or the unsloth/unsloth Docker image. Studio needs Python 3.11 to 3.13. Training runs on NVIDIA, AMD, Intel and Mac; CPU-only covers chat and Data Recipes. No usage telemetry per the docs. Your real cost is the GPU and the electricity.
Paid: unsloth Pro (pitched as 2.5x faster training, 20% less memory than the open source version, up to 8 GPUs) and unsloth Enterprise (multi-node, full training, faster inference, customer support), both 'Contact us' with no public price. The pricing page looks stale (it still lists multi-GPU as 'coming soon' for Free while the README says multi-GPU works). Licensing: Core is Apache 2.0 except the AGPL-3.0 MoE kernels in unsloth/kernels/moe; Studio (and the Desktop UI built on it) is AGPL-3.0.
Free for local training and inference on your own hardware. Pro and Enterprise are contact-sales only with no public price, and the AGPL parts (Studio UI, MoE kernels) matter only if you ship or host what you build on them.
What to do by team size
- Solo
- free; the Desktop app on your own GPU or Mac
- Small team
- free; the AGPL pieces only matter if you redistribute or host them
- Medium team
- free for most fine-tuning; ask sales about Pro only if multi-GPU speed is the bottleneck
- Large team
- Enterprise via sales for multi-node training and support; have legal review the AGPL components before building a product on it
Get tools like this every Wednesday
One featured tool, three on the radar. No fluff.
Similar Tools
Get up and running with Kimi-K2.5, GLM-5, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
Load almost any open AI model in a few lines of Python. The standard library for running and fine-tuning models from the Hugging Face Hub, now PyTorch only.
High-throughput LLM inference and serving engine
Open-source AI engine, run any model locally
SDK and proxy to call 100+ LLM APIs in OpenAI format
The Modular Platform (includes MAX & Mojo)
A low score is not a verdict on quality. Young and niche tools start low by design. How we calculate scores
Trust Signals
License: Apache-2.0 + AGPL-3.0 (Studio UI, MoE kernels)
Core library is Apache 2.0, but the Studio UI and the MoE kernels are AGPL-3.0, which matters if you ship or host a product on them.
Commercial use: ✓ Yes
About
- Owner
- Unsloth AI (Organization)
- Stars
- 79,099
- Forks
- 7,363
Also by Unsloth AI
Explore Further
More tools in the directory
rea
Reverse engineer anything with agents, from app behavior down to native binaries.
435.8k ★everything-claude-code
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
298.1k ★hermes-agent
The agent that grows with you
261.0k ★Featured in The Open Source Drop #22