Unsloth’s v0.1.915-beta gives local AI users a more capable built-in browser, faster MLX model loading and a collection of training improvements. The practical standout is that supported models can now read and edit files directly in the browser, while switching between GGUF models no longer means losing the chat context.
Unsloth Watch analysis
What happened
Unsloth published the beta release on 4 November. Its notes also describe lower idle memory use, faster MLX loading and reloading, and FP8 LoRA training with lower GPU memory use on recent NVIDIA GPUs. The release includes several changes to training support, browser tools and model connections, rather than one headline feature wearing a very large hat.
Our top picks
- Read and edit files in the browser
With browser use enabled, models can read and edit HTML, React, Markdown, code and text files; Back restores the previous version. - Keep chat context when switching GGUF models
Switching away and back preserves the conversation, avoiding the need to reread long histories. - Spend less memory while idle
Unsloth says it no longer loads PyTorch and other machine-learning libraries at startup. - Faster MLX loading and reloading
The notes say MLX models load and reload faster, and reducing a loaded model’s context now takes seconds. - More training options on MLX
DPO, ORPO, KTO, GRPO and CPO training now run through MLX. - Lower-memory FP8 LoRA training
The release adds FP8 LoRA training on 16-bit bases with recent NVIDIA GPUs.
Why it matters
These changes address familiar friction for people running models locally: waiting for models to load, losing context between models, and juggling separate tools to inspect or edit files. The training additions also broaden what users can run on their own hardware. That is a useful update for local AI workflows, not merely a fresh coat of glass-effect paint.
Our read
This beta is worth a look if you use Unsloth Studio for local models or training. The browser editing and retained context are the most immediately tangible upgrades; check the release notes for the hardware and feature conditions that apply to your setup before rearranging your workflow around them.
What to watch
- Whether the browser’s file-editing tools behave reliably on real projects.
- How much the lower idle memory use and faster MLX loading change everyday performance.
- Whether the new training paths work smoothly across supported models and hardware.
Discussion spark: Would you trust a local model to edit project files in its built-in browser, or keep that work in a separate tool you can inspect?
Sources and evidence
- Improved Customization + Glass + Browser Use (4 November 2026, 17:54 UTC)
- Releases · unslothai/unsloth · GitHub (4 November 2026, 17:54 UTC)
WittyWires independently tracks public Unsloth AI developments and is not affiliated with, endorsed by, or speaking for Unsloth AI, its maintainers, GitHub or X.