Unsloth AI
8 Intel
Unsloth AI is a San Francisco-based startup, founded in 2023 by brothers Daniel and Michael Han, that builds open-source tools for efficiently fine-tuning and training large language models. Daniel Han previously worked at NVIDIA optimizing algorithms and fixing bugs in open-source LLMs; Unsloth's custom GPU kernels cut memory use and training time significantly versus standard fine-tuning approaches. The company went through Y Combinator's Summer 2024 batch and the GitHub Accelerator, with backing from investors including Microsoft's M12 fund.
Intel
Unsloth brings local LLM training and inference to AMD GPUs across Windows, WSL and Linux, with custom Triton kernels claiming up to 2x speed and 70% less VRAM at no accuracy cost across Radeon, Instinct and Ryzen hardware.
Unsloth v0.1.913-beta adds DeepSeek-V4.1-Flash-1017 Dynamic v3.1 GGUFs in preview, Qwen-Image rendering up to 2x faster at 16 GB VRAM, batched MoE fine-tuning, faster app updates, Linux performance fixes, custom backgrounds and 14 new languages.
Unsloth v0.1.910-beta adds Flint, MOSS-Image and Nemotron 3.5 Bolt, SSD-backed experts for large MLX MoE models on Macs, REAP expert pruning, a chat-only desktop install and an OpenAI-compatible fine-tuning API.
Unsloth v0.1.915-beta adds Qwen3.8-Flash-Next dynamic quants, FP8 LoRA training with lower GPU memory, browser-use file interaction, faster MLX loading, preserved context when switching GGUF models, and Glass theme updates.
Unsloth v0.1.912-beta batches up to four compatible fine-tuning jobs with multi-LoRA, halves first-launch time, cuts Linux tool-call startup to 1 to 2 ms, extends thinking and auto-compact to SGLang, and adds a browser-use switch.
Unsloth v0.1.914-beta brings CPU text training, faster generation, more multi-GPU training options, chat navigation and Glass themes, plus an analysis of its DeepSeek-V4.1 GGUFs on agentic benchmarks.
Unsloth v0.1.911-beta adds scheduled prompt tasks on local or API models, MiniMax-M3.1-Flash support, FLUX 3 image editing, saved per-model run presets, and Mac performance gains. Scheduled tasks run only while the app is open and without tools or web search.
Unsloth's release adds local DeepSeek-V4.1-Flash-1017 and FLUX.3 Image Klein support, image and video chat, alpha FSDP2 multi-GPU training, token preview, and an MCP setup command for Claude Code, Codex and Cursor.