Unsloth just dropped Studio (Beta), and honestly? It’s the most ambitious local AI tool I’ve seen. Here’s my take after digging through the docs.

What Unsloth Studio Actually Does

Three things, but done well:

  1. Run GGUF models locally — Mac, Windows, Linux. No cloud.
  2. Train models 2x faster with 70% less VRAM
  3. Export anywhere — GGUF, safetensors, Ollama, vLLM

The Standout Features

Code Execution + Tool Calling

Most exciting part. Unlike chat interfaces that just talk, Unsloth Studio lets models:

  • Run Bash and Python code directly
  • Execute real computation to verify answers
  • Self-healing tool calling (it fixes broken calls automatically)
  • Web search inside the thinking trace

This is what Claude Artifacts wishes it could do locally.

No-Code Training

Upload a PDF, CSV, or JSON and start training. Fine-tune Qwen3.5, Nemotron 3, or 500+ other models. Multi-GPU works automatically.

Model Arena

Battle two models side-by-side. Compare your fine-tuned model vs base to see what changed. Simple but useful.

How It Compares

Feature Ollama LM Studio Unsloth Studio
Running ✅ ✅ ✅
Training ❌ ❌ ✅
Code execution ❌ ❌ ✅
Export Limited Limited Full
No-code UI Basic Polished Yes
Price Free Free Free

For My Setup

On M4 Max with 128GB:

  • Keep Ollama for quick tasks
  • Try Unsloth Studio for training
  • LM Studio for polished GUI when I just need to chat

Verdict

Unsloth Studio is the most complete local solution I’ve found. The code execution + training combo is unique. Worth a look if you’re serious about local AI.

Try it: unsloth.ai/studio

This is not a sponsored post. I actually read the docs.