Join waitlist
Back to products
Unsloth Studio

Unsloth Studio

Open-source web UI to run and train AI models. Open Source • Artificial Intelligence • Development 16 151 AssemblyAI Voice Agent API One API to build production-ready voice agents API • Artificial Intelligence • Audio

Website unsloth.ai
Overview

What it is

Open-source running and training of AI models and LLMs.

Intent

I need it when

Train custom LLMs faster with lower memory requirements on local hardware

Unsloth Studio enables 2x faster training with 70% less VRAM through optimized custom kernels supporting LoRA, FP8, and full fine-tuning across 500+ models. Users upload PDF/CSV/JSON datasets and train locally without cloud costs or data privacy concerns.

Transform unstructured documents into training datasets without manual annotation

Data Recipes feature auto-converts PDFs, CSV, JSON, DOCX files into usable datasets via graph-node workflow powered by NVIDIA Nemo Data Designer, eliminating manual data preparation overhead.

Export fine-tuned models for deployment across different inference stacks

Unsloth Studio exports trained models to GGUF or 16-bit Safetensors formats compatible with llama.cpp, vLLM, Ollama, and LM Studio, enabling flexible deployment without vendor lock-in.

Compare model outputs side-by-side to evaluate fine-tuning improvements

Model Arena feature lets users load two models simultaneously and chat with both to directly compare base model vs fine-tuned model outputs, with real-time observability tracking training loss and GPU utilization.

Run open-source models locally with advanced inference features like web search and code execution

Unsloth Studio runs GGUF and Safetensors models 100% offline on Mac/Windows/Linux with self-healing tool calling, web search, Python/Bash code execution, and OpenAI-compatible API endpoints for integration with Claude Code and other tools.

Drop

Not a fit when

  • User requires cloud-based inference without local hardware setup; Unsloth Studio runs 100% offline locally and requires compatible GPU or CPU
  • User needs proprietary closed-source model training; Unsloth is open-source and does not support proprietary model architectures outside its 500+ supported models
  • User lacks technical setup capability; Unsloth Studio requires installation via CLI commands and local environment configuration
  • User requires real-time collaborative multi-user training; Unsloth Studio is designed for single-machine local use without built-in multi-user collaboration
  • User needs AMD GPU training support; AMD chat inference works but training support is coming soon, not currently available
Commercials

Pricing

Freemium: Free tier with open-source standard version; Pro and Enterprise tiers available with contact-based pricing View pricing