Unsloth: baixar versions anteriors

This release adds scheduled tasks, FLUX-3 edit support, perf improvements and more. Also new: MiniMax-M3.1-Flash, face touch-ups for images and faster performance + chats on Mac.
Highlights
- Scheduled tasks: Run a prompt on a schedule with a local model or an API
- MiniMax-M3.1-Flash: MiniMax's 428B MoE (23B active) Guide
- FLUX 3 Editing: FLUX 3 now supports editing of images with prompts
- - Perf improvements: Large Mac perf gains, less lag, reduced binary sizes
- Saved presets: Save your Run settings as named presets for each model
Scheduled tasks
- Open Scheduled from the sidebar's More menu to run a prompt once, daily, on weekdays, weekly or monthly; each answer is saved as a new chat.
- Each task can run on the loaded model, a model on your device, or a model from a connected API provider.
- When a task's model isn't loaded, Unsloth asks before loading it, unless you turn on Switch automatically when I'm away.
- Each task can notify you after every run, only when it fails, or never.
- Tasks run only while Unsloth is running, without tools or web search.
UI + UX
- Save Run settings in the model picker as named presets for each model, then rename, delete or switch between them.
- The preset you picked last is what that model loads with next time, including loads through the API.
- Chat settings can remember your sampling settings for a model, or reset them to the model's recommended values.
- GGUF chats now show progress and time left while processing a long prompt, before the reply starts.
- The API page now shows each served model's context use, speed and request count.
- Each kind of button and icon now looks the same everywhere in the app.
- The Library and System pages in Settings show storage as a usage bar, split into models, caches and free space.
Mac
- MLX chats now reply up to 16% faster with draft models.
- MLX models that ship with a built-in draft model now use it automatically, with no separate download.
- Image and video renders need far less memory at the end, so low-memory Macs rarely slow down from swapping.
- On Apple Silicon, Z-Image GGUFs now make images from text with the MLX engine by default, 1.5-2x faster than before.
Images + video
- Make videos from text or an image with NVIDIA's Cosmos3-Edge on the Video page.
- On the Images page, turn on Detailer to redraw faces, hands or other things you name at higher resolution; it needs SAM 3 and a model that can inpaint.
- On NVIDIA GPUs older than RTX 30-series, Wan2.1 1.3B text-to-video clips render about 5x faster.
- On NVIDIA GPUs, many image and video models load faster after the first time, since Unsloth saves the version it builds on first load.
- FLUX.3 Klein can now edit photos in Images > Edit, keeping the photo's dimensions.
Training
- Split each layer of a model across GPUs for LoRA and QLoRA training, so each GPU holds less: set UNSLOTH_TP_SIZE to the number of GPUs and launch with torchrun.
- Fine-tuning Unsloth's 4-bit gpt-oss checkpoints no longer crashes on the first training step.
- GRPO that generates with vLLM runs up to 1.9x faster per step with 16-bit LoRA on Linux with NVIDIA GPUs.
- GRPO that generates without vLLM runs up to 4.4x faster per step for Llama and Qwen3 on NVIDIA GPUs.
Coding agents + API
- Apps using an Unsloth API key can no longer approve their own Python or terminal calls to run outside the sandbox.
- Unsloth no longer trusts forwarded client IPs by default; behind your own reverse proxy, set UNSLOTH_STUDIO_TRUST_FORWARDED=1 or all users share one login rate limit.
Models + hardware
- Run MiniMax-M3.1-Flash, MiniMax's MoE model for coding, tool use and agent workflows, with up to 1M tokens of context. GGUF
- At 1-bit, MiniMax-M3.1-Flash needs at least 130 GB of RAM and VRAM combined.
- On some Windows PCs, GGUF models load again instead of failing every time with "llama-server failed to start".

Recursive's Flint, MOSS-Image and NVIDIA's Nemotron Bolt 4B are here! Run them locally in Unsloth. Also new: a chat-only Unsloth Desktop install, MLX MoE models bigger than your Mac's memory and an OpenAI-compatible fine-tuning API.
Highlights
- Bigger MoE models on Mac: MLX MoE models too big for memory keep experts on SSD, so Qwen3.8-Flash-Next runs on a 32 GB Mac.
- Chat-only install: a much smaller Unsloth Desktop for GGUF chat; add training later in Settings.
- Flint: Recursive's 40B coding-agent MoE, runs in 24 GB at 3-bit Guide
- MOSS-Image: 5.1B image model designed for top-tier generation GGUF + FP8
- NVIDIA Nemotron 3.5 Bolt E4B: NVIDIA's new Gemma 4 E4B fine-tune, 63.2 on Tau2-Bench vs the original's 38.7 Guide
- MoE expert pruning: Export a smaller MoE model by removing 25% or 50% of its least useful experts with REAP.
- UI overhaul: More consistent UI across Unsloth
Images + video
- Generate images containing English or Chinese text with OpenMOSS's 5.1B MOSS-Image, SFT or Base. GGUF · Base GGUF
- MOSS-Image, the new image model, runs on 16 GB GPUs; its GGUFs also run on Macs and computers without a supported GPU.
- Most image and video models too big for one GPU now split across your NVIDIA GPUs, if they fit, instead of streaming from system memory.
- The new Hires fix upscales and refines a render in one step on SDXL, FLUX.1, Qwen-Image and Z-Image.
- The new Compare grid renders one prompt while varying up to three settings, such as seed, CFG or LoRA weight, in labelled grids.
- SDXL safetensors get AUTOMATIC1111-style prompt weighting, CLIP skip, CFG rescale, variation seeds and textual inversion.
- Pasted AUTOMATIC1111, Forge or Civitai parameter blocks fill in the Images page settings.
- If SAM 3 isn't already downloaded, the Mask from text tool now uses SAM 3.1: smaller, with no Hugging Face login.
Chat + Desktop
- Pick Chat only (GGUF) when first launching Unsloth Desktop for a much smaller install; add training later in Settings.
- Settings > Chat now lets you view, edit, clear or turn off the memory notes Unsloth sends to every chat.
- In document-search chats, Show passages lists the exact text the model got and opens each passage at its page.
- The Audio page can speak and transcribe via a saved API connection, such as OpenAI or a local server.
- Unsloth installs, updates and pip install -U unsloth now reject eMudhra's compromised emSign certificates, which let attackers impersonate websites. The optional vLLM and SGLang engines still trust them.
Flint
- Run Recursive's Flint, a 40B MoE model with 2.5B active parameters for coding agents, with reasoning, tool use and 256K context. Guide · GGUF
- At 3-bit, Flint fits in 24 GB at short contexts.
- On SWE-bench Pro v2, Flint scores 89.7 vs Qwen 3.6's 73.8.
NVIDIA Nemotron 3.5 Bolt E4B
- Run and fine-tune NVIDIA's Gemma Nemotron 4B, a Gemma 4 E4B reasoning tune for tool calling and instruction following.
- Gemma Nemotron 4B beats Gemma 4 E4B on 16 of 21 NVIDIA tests, gaining most in tool use and multi-turn chat.
- DPO on Gemma 4 E2B and E4B, including Gemma Nemotron 4B, no longer crashes.
Training
- Fine-tuning Whisper, Canary, Moonshine, Dia and CTC speech models with FastModel no longer fails.
- Parakeet, Canary and CTC speech training uses up to 55% less memory.
- The new Books & Documents to Text recipe in Data Recipes turns ebooks and documents into a pretraining text dataset, no model needed.
- On Windows and Linux, the Export page can now prune 25% or 50% of a MoE model's least useful experts with REAP, at some quality cost.
- In Python, reap_prune runs REAP expert pruning up to 54x faster than llm-compressor's.
- unsloth train now trains vision and, except on Mac, audio models.
- Unsloth now supports OpenAI's fine-tuning API, so OpenAI's SDK can upload datasets and run supervised fine-tuning of local models.
Performance
- MLX MoE models too big for your Mac keep experts on SSD, with slower replies. Qwen3.8-Flash-Next fits a 32 GB Mac; DeepSeek-V4-Fl

DeepSeek-V4.1-Flash-1017 and FLUX.3 Image Klein are here! Run both locally in Unsloth. Also new: images and videos in chat, multi-GPU training with FSDP2 (alpha) and one-command MCP setup for coding agents.
Highlights
- DeepSeek-V4.1-Flash-1017: 552B (16B active) - runs on ~125 GB at 1-bit with its Engram table on SSD Guide
- FLUX.3 Image Klein: Runs on 24 GB and edits with up to 10 reference images GGUF
- One-command MCP setup: unsloth mcp add connects Claude Code, Codex or Cursor to Unsloth.
- Images and videos in chat: Local chat models use the image or video model loaded beside them.
- Qwen-Image-Edit-2509 and 2511: Edit with up to 3 images at once.
- Token preview: See what your model will learn before training starts.
DeepSeek-V4.1
- Run DeepSeek-V4.1-Flash-1017, an update to DeepSeek-V4.1-Flash with better reasoning, coding and agentic results, and shorter outputs. It has 552B parameters plus 196B in its Engram table, about 16B active, and 1M tokens of context. With the Engram table left on SSD, it runs in about 125 GB of memory at 1-bit, or 300 GB at 4-bit. Guide · GGUF
- DeepSeek-V4.1-Flash-1017 scores 42.3 on Terminal-Bench 4.0, up from 31.2 for the earlier DeepSeek-V4.1-Flash.
- Thinking is on by default at high effort. Choose any effort from low to max, or turn thinking off.
FLUX.3 Image Klein
- Generate and edit images with Black Forest Labs' FLUX.3 Image Klein, a 5.6B model that renders text and edits with up to 10 reference images. It runs on 24 GB of GPU memory, its GGUFs also run on computers without a supported GPU, and the undistilled Klein Base is available too. GGUF · Base GGUF
- Enhance prompt, on by default, expands short prompts into detailed captions, which fixes most distorted bodies in hard poses. Preview shows the caption so you can copy or edit it.
Images + video
- Qwen-Image-Edit-2509 and Qwen-Image-Edit-2511 now accept up to 3 images per edit.
- Set custom video sizes and aspect ratios in Create videos, or pick one of the model's trained sizes.
- Start an edit straight from an image's menu in the Images gallery or the Library, or drag a generated image onto an image picker on the Images page.
- Image and video models with a large text encoder now load a quantized encoder by default where the GPU supports it, cutting the encoder's memory by up to half and loading faster. LTX-2.5 no longer needs a Hugging Face login.
Chat + Desktop
- Image, video and Read aloud speech models can now stay loaded beside your chat model on computers with a dedicated GPU. Pick When they fit or Always in Settings > Resources > Model memory.
- With the Images pill on, a local chat model can use the image or video model loaded beside it to make images and playable videos in the chat.
- After an image or video model replaces your GGUF chat model, the chat continues without re-reading the conversation, as long as Keep chat context across unloads is on.
- In Unsloth Desktop, choose an existing working folder when creating a project.
- Unsloth Desktop no longer crashes at startup on older macOS versions, and its browser panel works on macOS 27 and follows system proxies such as Clash.
Training
- In Python, spread LoRA, QLoRA and full fine-tuning across GPUs with FSDP2, and split long sequences, including packed ones, with context_parallel_size. Check the machine first with unsloth doctor --gpus.
- On the Train page, Preview tokens shows which parts of your dataset the model trains on and flags samples cut off by the context limit. Data Recipe chats now train as multi-turn conversations, and Qwen fine-tunes keep a dataset's thinking text.
- Cut full fine-tuning optimizer memory with optim = "adamw_flash", or set UNSLOTH_OFFLOAD_OPTIMIZER_STATE=1 to keep optimizer state in system memory between steps at some cost in speed.
Performance
- Local GGUF chats with tools now show replies as fast as the model writes them, so the next tool call no longer waits, and long replies no longer slow down as they grow.
- LoRA training with lora

Run NVIDIA's new Nemotron 3.5 models and Meta's new Muse Spark Open locally in Unsloth. New: text-diffusion LLMs, DeepSpeed training and better prompt caching for coding agents.
Highlights
- NVIDIA Nemotron 3.5 Bolt: Run on 24 GB of memory. Guide
- NVIDIA Nemotron 3.5 Super VL: 120B (12B active) Guide
- Meta Muse Spark Open: 816B (42.7B active) - runs on 256GB Guide
- Better prompt caching for coding agents: Claude Code, Codex etc. on local GGUF models re-process far less of each prompt.
- Browser use: Always allow properly works, cursor, and separate tabs for each chat.
- Images through your connections: generate with sd-server, Ollama, Lemonade and other image servers.
- Diffusion LLMs: train and chat with LLaDA, Dream, SDAR and more, with a live denoising canvas.
Nemotron 3.5
- Run NVIDIA Nemotron 3.5 Super VL, a 120B model with 12B active parameters and up to 1M tokens of context, built for document understanding, computer use and long-running agents. Super VL runs in about 64 GB of memory at 1-bit, or 128 GB at 4-bit. Guide · GGUF
- Run NVIDIA Nemotron 3.5 Bolt, a 74B model with 9B active parameters and 128K context, compressed from Super VL. Bolt runs in about 32 GB of memory at 1-bit or 2-bit. Guide · GGUF
- Nemotron 3.5 Super VL generates about 5x faster than Inkling Small at 100K input with similar intelligence.
- By default, Unsloth runs Nemotron 3.5 Bolt with a DSpark drafter (about 1 to 2 GB more memory) and Super VL with its built-in MTP for faster output. Both can be turned off in settings. MTP Guide
Muse Spark Open
- Run Meta's Muse Spark Open, an 816B MoE model with about 42.7B active parameters and 1M tokens of context, built for multimodal reasoning, coding and agentic workflows. Muse Spark Open runs in about 256 GB of memory at 2-bit. Guide · GGUF
- Muse Spark Open takes text, images, video and audio as input, with reasoning effort from low to xhigh.
- Muse Spark Open was trained on long-horizon coding tasks, including whole-repository generation, large end-to-end projects and auto-research.
Training
- Train and chat with text-diffusion LLMs in the Unsloth app, including LLaDA, Dream, SDAR, LLaDA2, Nemotron-Labs-Diffusion and DiffusionGemma. Chat replies show a live denoising canvas.
- Train with DeepSpeed ZeRO-2 and ZeRO-3, including 4-bit QLoRA, full fine-tuning, DPO and CPU offload. LoRA training with DDP and gradient checkpointing also no longer crashes.
- GRPO with fast_inference now runs on Nemotron 3 Nano and block-FP8 MoE models, sharing weights with vLLM.
- On the Train page, stop early when eval loss stops improving and train on pre-tokenized datasets. Random Seed now also sets the LoRA starting weights.
Performance
- When Claude Code, Codex or OpenCode runs on a local GGUF model, follow-up turns re-process much less of the conversation, and a new Claude Code session's first reply starts up to 4x sooner.
- In Python, single-prompt model.generate is up to 2.7x faster at 32K context for Llama and Qwen3 models.
- In Python, generating 2 to 4 prompts at once with a bitsandbytes 4-bit model is up to 2.8x faster on some NVIDIA GPUs.
- GRPO rollouts with fast_inference on Nemotron 3 Nano are up to 30% faster by default.
Browser use
- See a cursor where the model clicks and types, or turn it off in Settings > Browser > Browser use > Show cursor.
- Browser use prompts now have Always allow, which approves that kind of action on a site for the rest of the chat. Scripts still ask every time.
- Each chat keeps its own browser tabs, and in Desktop, browser use keeps working in chats that are not on screen.
- In the web UI, browser pages remember site preferences between pages, load data from their own site, and play videos full screen.
Images + video
- Generate images on the Images page through a connected image server, such as sd-server, Ollama, Lemonade, vLLM, llama.cpp or a custom OpenAI-compatible URL.
- Image and video models now skip text-encoder layers they never use, saving memory on every load.
Chat + 

Turn any text or vision LLM into a Jev-style decision model in Unsloth, with decision accuracy going from 30% to 80%. Train, test, export and serve decision models directly from Unsloth. Also included: native ComfyUI models, diffusion improvements and a better Browser in Desktop.
Highlights
- 8th Oct Fixes - Better Browser + 100+ bug fixes + perf fixes
- Sandboxing with Bwrap for Linux, Seatbelt for Mac and MXC for Windows
- Turn any model into a Jev-style decision model. Accuracy went from 30% to 80%
- Load ComfyUI diffusion models natively in Unsloth
- Faster + more accurate diffusion with INT8 ConvRot and more
https://github.com/user-attachments/assets/92959172-d2a2-450b-a5bf-6369a1e5ae16
Decision models
- Train any text or vision LLM as a Jev-style decision model using QLoRA.
- Test trained models directly from the Decision API settings.
- Make decisions with confidence scores for every option.
- Export Clef models with Qwen3.5 backbones and Laya models to GGUF.
- Serve supported decision models through llama.cpp, including models that understand images.
- Save and resume smaller adapter and decision-head checkpoints.
- Guide at https://unsloth.ai/docs/basics/train-your-own-decision-model-with-unsloth
Diffusion + ComfyUI
- Run supported ComfyUI image and video models directly from Hugging Face.
- Unsloth now recognises ComfyUI checkpoints and local model folders automatically.
- Use your ComfyUI text encoders and VAEs in Unsloth.
- Run Krea-2, HunyuanImage-2.1 and Wan2.2 expert pairs, plus ComfyUI NVFP4 and MXFP8 models.
- Qwen-Image-2.1 now keeps more full-precision image detail with INT8 ConvRot enabled by default.
- Faster Qwen-Image-2.1 ConvRot generation on supported NVIDIA GPUs.
Training + performance
- Train Qwen3.5-35B-A3B up to 4.1x faster and Qwen3-30B-A3B up to 3.3x faster with QLoRA on A100 and RTX PRO 6000.
- Improved sample packing for gated-delta, Mamba2 and short-convolution models.
- Train prompt and completion message lists as one conversation.
- Vision datasets now keep each row's own question.
- Chat exports and training data now include the system prompt.
Browser + Desktop
- Ask about open pages in the Desktop Browser.
- Confirm Browser downloads and choose where files are saved.
- Reorder pinned pages in the sidebar like chats.
- Search continues past unusable results and can fall back to Wikipedia.
- Reply citations such as [1] now open as links.
- Pick, pin or unload RAG embedding models directly from the RAG menu.
Sandboxing
- Bwrap on Linux, Seatbelt on Mac and MXC on Windows sandbox code the model runs.
- View sandbox status and choose protection levels in Settings.

Turn any text or vision LLM into a Jev-style decision model in Unsloth, with decision accuracy going from 30% to 80%. Train, test, export and serve decision models directly from Unsloth. Also included: native ComfyUI models, diffusion improvements and a better Browser in Desktop.
Highlights
- Turn any model into a Jev-style decision model. Accuracy 30% to 80%
- Load ComfyUI diffusion models natively in Unsloth
- Faster + more accurate diffusion with INT8 ConvRot and more
- Better Browser in Desktop with many bug fixes
- Sandboxing with Bwrap for Linux, Seatbelt for Mac and MXC for Windows
https://github.com/user-attachments/assets/92959172-d2a2-450b-a5bf-6369a1e5ae16
Decision models
- Train any text or vision LLM as a Jev-style decision model using QLoRA.
- Test trained models directly from the Decision API settings.
- Make decisions with confidence scores for every option.
- Export Clef models with Qwen3.5 backbones and Laya models to GGUF.
- Serve supported decision models through llama.cpp, including models that understand images.
- Save and resume smaller adapter and decision-head checkpoints.
- Guide at https://unsloth.ai/docs/basics/train-your-own-decision-model-with-unsloth
Diffusion + ComfyUI
- Run supported ComfyUI image and video models directly from Hugging Face.
- Unsloth now recognises ComfyUI checkpoints and local model folders automatically.
- Use your ComfyUI text encoders and VAEs in Unsloth.
- Run Krea-2, HunyuanImage-2.1 and Wan2.2 expert pairs, plus ComfyUI NVFP4 and MXFP8 models.
- Qwen-Image-2.1 now keeps more full-precision image detail with INT8 ConvRot enabled by default.
- Faster Qwen-Image-2.1 ConvRot generation on supported NVIDIA GPUs.
Training + performance
- Train Qwen3.5-35B-A3B up to 4.1x faster and Qwen3-30B-A3B up to 3.3x faster with QLoRA on A100 and RTX PRO 6000.
- Improved sample packing for gated-delta, Mamba2 and short-convolution models.
- Train prompt and completion message lists as one conversation.
- Vision datasets now keep each row's own question.
- Chat exports and training data now include the system prompt.
Browser + Desktop
- Ask about open pages in the Desktop Browser.
- Confirm Browser downloads and choose where files are saved.
- Reorder pinned pages in the sidebar like chats.
- Search continues past unusable results and can fall back to Wikipedia.
- Reply citations such as [1] now open as links.
- Pick, pin or unload RAG embedding models directly from the RAG menu.
Sandboxing
- Bwrap on Linux, Seatbelt on Mac and MXC on Windows sandbox code the model runs.
- View sandbox status and choose protection levels in Settings.

This release brings faster navigation, shareable run settings, and clearer errors to Unsloth Desktop. It also 4x speeds up Laya decisions, expands hosted Decision API support, and keeps NVFP4, INT4, and MXFP4 checkpoints in 4-bit during LoRA training.
Highlights
- Command palette on Cmd/Ctrl+P. Also, you can now share your GGUF model's run settings
- Laya decisions up to 4.1x faster, plus connections to TypeSafe and other hosted providers
- Clearer errors that explain failed loads and generations, with a View logs button
- NVFP4 LoRA with 45% lower peak memory on Qwen3.8-27B-NVFP4
- INT4 + MXFP4 checkpoints stay packed during LoRA, including Kimi-K2.7-Code and gpt-oss
Desktop + chat
- Cmd/Ctrl+P opens a command palette, and GGUF run settings can be shared as a link or saved without loading.
- Continue response extends a finished reply, and Fix with the model puts HTML canvas errors in the message box.
- A warning flags models that only partly fit on the GPU, and the Images gallery loads thumbnails first.
- The desktop app adds Cmd/Ctrl + and - zoom, Windows 11 style buttons on Windows and Linux, and a check before Update stops training.
Lower-memory checkpoint training
Pre-quantized 4-bit checkpoints now train on their exact published weights. Training reads them packed, so a loaded model needs about as much memory as its 4-bit checkpoint.
- Qwen3.8-27B-NVFP4 peaks at 40.2 GB instead of 72.9 GB on one RTX PRO 6000, with 11% shorter steps.
- Dense compressed-tensors INT4 checkpoints train as published on NVIDIA GPUs. Qwen3-8B w4a16 peaks at 8.5 GB, not 18.8 GB, on a B200.
- Kimi-K2.7-Code trains on 6 B200s with its INT4 experts packed, where a 16-bit copy would need about 2 TB.
- gpt-oss loaded with load_in_4bit = False keeps its MXFP4 experts, so gpt-oss-120b trains on one B200.
How it works
- Triton kernels dequantize each layer on the fly, bit for bit matching compressed-tensors, and backward keeps only the packed weights.
- INT4 stays exact instead of being re-quantized to NF4, which rounds every weight twice and raised Qwen3-8B perplexity by 1.6%.
- For inference-only gpt-oss, UNSLOTH_MXFP4_KEEP_PACKED=0 restores the native load, which prefills faster.
Decision API + Laya
Laya decision models now answer short requests up to 4.1x faster, and hosted decision models work too.
- Add TypeSafe or another hosted decision provider under Connections, then pick its models in Settings > API.
- With the Decision API on, enable Unsloth Decisions in the chat's MCP menu for a decide tool.
- On a B200, a 3-question guardrail check takes 2.8 ms instead of 11.6 ms. Long inputs run at about the same speed.
- Laya loads in about 9 s instead of 17 s and needs about half the peak host RAM.
Image + video
- Image models keep INT8 or FP8 when offloading. On a 10 GiB-capped B200, Z-Image takes 2.3 s, not 38.4 s.
- Qwen-Image-2.1 renders 1536x1536 images up to 3.6x faster on a Radeon 8060S.
Training + models
- Fine-tune T5, T5Gemma, BART, Marian and other text encoder-decoders with FastModel.
- The new unsloth eval command runs lm-evaluation-harness tasks on a checkpoint or LoRA adapter.
- Gemma 4 31B training and batched generation now work on models split across GPUs.
AMD + Apple Silicon
- On Windows, RX 5000 through RX 9000 cards get PyTorch from AMD's multi-arch index.
- MLX moves to mlx 0.32.3 and mlx-vlm 0.7.4, and quantized-KV vision loads can batch replies.
- On a Mac, export LoRA adapters as PEFT or GGUF for GPU machines.

This is a large performance and reliability + bug fix release for Unsloth
Highlights
- 1.2-1.7x faster diffusion. AMD 20% perf boost vs ROCM via Vulkan
- 2x faster updating, remove SAC + AV false positives for Windows
- Blender MCP, detect Hermes, AMD gibberish fixed (reported to AMD)
- Over 250+ bug fixes, 60% smaller binaries and performance improvements
- Strix iGPU BIOS popup - 3x faster inference if more VRAM for iGPU
- Updated PyTorch to 2.11 from 2.10 - 2.14 will be soon. MLX improvements
- Default to gpt-6-astra for Codex logins and improve OpenAI API processing
- MLX fixed when self healing / updating, causing slowdowns for inf + training
- Fix AppImage being very laggy. Fixed RAG upload issues stuck at 28% and API issues
- New Docker image published at https://hub.docker.com/r/unsloth/unsloth for Studio and notebooks
Performance Boosts
- Diffusion is 1.2x to 1.7x faster for INT8 / FP8 pathways - all models accelerated.
- 23% faster prompt processing and 8% faster generation on Strix Halo.
- Gated-delta models now train up to 25% faster on Apple Silicon.
- Quantized MLX KV caches use up to 74% less prompt memory than before.
AMD + Windows
- Strix Halo and Strix Point now default to Vulkan for faster inference.
- AMD iGPUs without ROCm now use Vulkan instead of CPU on Linux.
- Windows llama.cpp binaries are now signed to reduce Smart App Control blocks.
- Windows now clearly explains when code integrity blocks model loading.
- Reinstalling Unsloth on Windows keeps your supported PyTorch version.
MLX + Apple Silicon
- Fresh Mac installs keep MLX training and exports working.
- Fine-tune with DoRA and more DPO loss types on Apple Silicon.
- Batched MLX generation now streams and samples each chat independently.
- More multimodal models can be fine-tuned using text-only datasets.
Studio + API
- RAG and document uploads are faster, show clearer progress and no longer appear stuck.
- Studio menus, Find and Settings now open more smoothly.
- GPT-6 Astra is available for Codex logins with Low through Max reasoning controls.
- Improved OpenAI-compatible API streaming, audio input and model loading.
- AppImage builds are more reliable with a pinned release toolchain.
Installs + Docker
- PyTorch 2.11 is now the default across supported installers.
- New NVIDIA Docker images for training and Studio, from Turing to Blackwell.
- Native AMD64 and ARM64 images are available from Docker Hub.
- The Unsloth Python package is now >60% smaller.
Docker: https://hub.docker.com/r/unsloth/unsloth

This is a large performance and reliability + bug fix release for Unsloth
Highlights
- AMD uses Vulkan by default - 20% perf boost for prefill, decoding vs ROCM
- Windows llama-server.exe is now signed, reducing false positives for SAC
- AMD gibberish issues in Strix, iGPUs fixed in (upstream - reported to AMD)
- Over 200+ bug fixes, 50% smaller binaries and performance improvements
- Updated PyTorch to 2.11 from 2.10 - 2.14 will be soon. MLX improvements
- Default to gpt-6-astra for Codex logins and improve OpenAI API processing
- MLX fixed when self healing / updating, causing slowdowns for inf + training
- Fix AppImage being very laggy. Fixed RAG upload issues stuck at 28% and API issues
- New Docker image published at https://hub.docker.com/r/unsloth/unsloth for Studio and notebooks
AMD + Windows
- Strix Halo and Strix Point now default to Vulkan for faster inference.
- Up to 23% faster prompt processing and 8% faster generation on Strix Halo.
- AMD iGPUs without ROCm now use Vulkan instead of CPU on Linux.
- Windows llama.cpp binaries are now signed to reduce Smart App Control blocks.
- Windows now clearly explains when code integrity blocks model loading.
- Reinstalling Unsloth on Windows keeps your supported PyTorch version.
MLX + Apple Silicon
- Gated-delta models now train up to 25% faster on Apple Silicon.
- Quantized MLX KV caches use up to 74% less prompt memory than before.
- Fresh Mac installs keep MLX training and exports working.
- Fine-tune with DoRA and more DPO loss types on Apple Silicon.
- Batched MLX generation now streams and samples each chat independently.
- More multimodal models can be fine-tuned using text-only datasets.
Studio + API
- RAG and document uploads are faster, show clearer progress and no longer appear stuck.
- Studio menus, Find and Settings now open more smoothly.
- GPT-6 Astra is available for Codex logins with Low through Max reasoning controls.
- Improved OpenAI-compatible API streaming, audio input and model loading.
- AppImage builds are more reliable with a pinned release toolchain.
Installs + Docker
- PyTorch 2.11 is now the default across supported installers.
- New NVIDIA Docker images for training and Studio, from Turing to Blackwell.
- Native AMD64 and ARM64 images are available from Docker Hub.
- The Unsloth Python package is now 44% smaller.
Docker: https://hub.docker.com/r/unsloth/unsloth

Run Qwen3.8-Flash-Next and GLM-5.3-Flash up to 2x faster with MTP. MTP is enabled by default, you can still disable it.
Also our new release includes 170+ training, chat, hardware, and performance improvements.
Highlights
- Smoother model loading (less errors) across local servers and connected providers.
- Safer chat edits that preserve tool cards, reply details, and conversation branches.
- New local media APIs for video, audio, and MLX-served models.
- New audio support with new models, progress tracking including: MiniMax-Music3, Higgs, MOSS and more!
- Improved multi-GPU planning, memory fitting, and split-model training.
- Strengthened AMD/ROCm detection, installation, and GPU compatibility.
- Upgraded MCP, Deep Research, OAuth, and agent tool reliability.
Qwen3.8-Flash + GLM-5.3-Flash
- Qwen and GLM now generate faster with MTP enabled by default.
- Use GLM tools across longer, multi-turn chats.
- Qwen automatically applies the recommended settings for thinking and non-thinking modes.
Download Qwen3.8-Flash-Next and GLM-5.3-Flash. See the Qwen guide and GLM guide for recommended settings and available GGUFs.
Faster MLX inference
- Fine-tune both large MoE models with text or images on Apple Silicon using MLX.
- Long Qwen chats now run much faster on Mac, with follow-up turns up to 30x faster.
- MLX models now use their full context size and support much longer batched generation.
- MLX releases GPU memory more cleanly between generation bursts and model switches.
- Serve MLX models through Unsloth's OpenAI-compatible API.
Audio
- Added support for MiniMax-Music3, Higgs, MOSS audio models.
- Added live progress updates while audio is being generated.
- Audio clips can now be archived and managed.
- Improved reliability with custom TTS playback fixes, Whisper pairing checks, and stronger audio testing.
Chat + tools
- Run several tool calls at once without mixing up their arguments.
- Keep tools available when chatting with images.
- Each chat keeps its MCP connection for faster tool calls.
- Local models can edit code using Codex’s apply_patch tool.
- Continue long chats with images and other media using Auto Compaction.
- Review and approve Deep Research plans before research starts.
Training + hardware
- Train larger models across multiple GPUs with automatic placement.
- AMD installs choose the best build across Windows and Linux, with BF16 on more GPUs.
- Export GLM-5.3 MLX fine-tunes to GGUF.
- Choose custom GGUF shard sizes and save locations.
API + Desktop
- Generate videos through the new OpenAI-compatible Videos API.
- Updates download in the background and install when you restart.
- Choose a custom port for LAN access.
- Generate audio with Higgs, MOSS and MiniMax models.
- Track audio generation progress and archive finished clips.
- Model downloads show clearer progress and can switch from Xet to HTTP automatically.

Qwen3.8-27B and Qwen3.8-2.4T can now be run locally in Unsloth!
Run on 17GB RAM via Unsloth Dynamic GGUFs. You can also fine-tune Qwen3.8-27B in Unsloth.
Qwen3.8-27B is by far the strongest model for its size. We also uploaded NVFP4 quants.
Guide: https://unsloth.ai/docs/models/qwen3.8
GGUF: https://huggingface.co/unsloth/Qwen3.8-27B-GGUF
See 1-bit Qwen3.8-2.4T GGUF running in Unsloth:
Highlights
- Qwen3.8-27B + extra llama-server arguments allowed + custom VRAM toggle
- External provider has tool calling + tool support + login with Codex
- Fast FP8 10x faster MiniMax-H3 inference (3 minutes vs 30)
- 10% faster inference for GGUFs + Bypass permissions fixed
Chat + tools
- Connected AI providers can use their own Search or Unsloth Desktop's built-in Search and tools. Tool results are passed back to the model so it can continue multi-step tasks.
- Sign in with a Codex subscription and use Codex tools inside Chat.
- Chat shows live prompt and generation speeds, while long streaming replies use much less CPU.
- Chat settings stay with the conversation across remote sessions.
- Paste a YouTube link to attach its transcript, including the title, channel, duration, link, and caption language.
- Save a full chat or reply into your project's sources while keeping its reasoning, tool calls, and citations.
MiniMax-H3
- MiniMax-H3 can run on smaller supported GPUs by splitting large model parts into pieces that fit.
- The model picker now hides H3 options that the current hardware cannot run instead of letting them fail after selection.
- H3 options are labelled Fast FP8 or Slow, making the large speed difference clear before downloading.
Performance + hardware
- Inference is up to 10% faster in supported cases, with lower VRAM use and a tunable memory limit.
- Idle image and video models can optionally unload to free VRAM for Chat or Training.
- Added better support for AMD RDNA 3, RDNA 4, and Strix Halo systems. VRAM checks no longer reserve extra GPU memory.
- Multi-GPU ROCm device matching is safer.
- Macs now choose context size from the memory that is actually free.
- RAG document indexing uses the CPU by default, so it no longer leaves a large GPU memory block reserved.
- Fixed GGUF image detection when choosing a model for the API.
Custom llama.cpp arguments
- Model settings now include an Extra Arguments box for custom llama-server flags.
- Unsloth checks flags against the installed build and saves valid ones per model for normal, startup, and API loads. Flags that could break model loading or app security are rejected with a clear message.
Training + exports
- Training checks whether a model needs a newer Transformers version and asks before installing it. The preview also tells you when the model must use 16-bit instead of 4-bit.
- GGUF exports check disk space before starting a long merge, with clearer errors showing how much space is needed.
- On Kaggle, large saves can use the bigger temporary drive when the working folder is too small.
- Fixed LoRA being ignored by vLLM GRPO rollouts on newer TRL versions.
Debugging + reliability
- Settings > Debugging now shows the main server and model-runner logs, with live, 3-second, and manual refresh modes.
- Copy the visible log or its full path for a bug report. Passwords, tokens, and API keys are hidden in the viewer, while files on disk stay unchanged.
- Cached RAG models load first, avoiding long stalls when Hugging Face is unreachable.
- Slow installs can keep running during large downloads and show what they are waiting for.
- Fixed Hugging Face dataset links and linked folders on Windows.
- Pinned models can be reordered by dragging them in the model list.

Unsloth Desktop is here! The first desktop app to run and train AI models locally. Research, export and deploy from the same open-source app on Windows, macOS and Linux.
Here's what you can do with Unsloth Desktop:
- Get up to 50% more accurate tool calling with self-healing calls and sandboxed code execution.
- Run Muse Glimmer 30B, Kimi K3, Qwen3.8, DeepSeek-V4 Flash 0731, Gemma 4, and more.
- Generate videos with MiniMax-H3, and create images and videos with other diffusion models at up to 2× faster inference on supported workflows.
- Use unlimited private web search, Deep Research, RAG and MCP.
- Export models to NVFP4, GGUF and other formats.
- Access Unsloth remotely through Cloudflare HTTPS.
- Run on CPU or multiple GPUs across NVIDIA, AMD, Intel and Mac.
- Train models without code, using less time and VRAM.
- Use local models through Unsloth's OpenAI-compatible API, or connect OpenAI and Anthropic models.
Tools, private research + APIs
Self-healing tool calling repairs malformed calls instead of dropping them. Models can run Python and Bash inside sandboxed environments, so they can test code, create files and verify their work.
Use unlimited private web search, let Deep Research plan and produce cited reports, or bring your own files into RAG. You can also connect MCP tools for workflows that need external apps, data or actions.
Local models can be served through Unsloth's OpenAI-compatible API for agents and other clients. Inside Desktop, you can also connect OpenAI and Anthropic as cloud model providers.
Muse Glimmer 30B + latest models
Run Muse Glimmer 30B locally for chat, agents, tools and APIs, alongside Kimi K3, Qwen3.8, DeepSeek-V4 Flash 0731 and Gemma 4. Download and manage them in one place through Unsloth Desktop.
MiniMax-H3 + image and video diffusion
Run MiniMax-H3 locally for video generation. Create images and videos locally, edit existing images and train supported diffusion models. Use LoRAs, reference images and ControlNet where available, with up to 2× faster inference on supported workflows.
No-code training, export + remote deployment
Pick a model and dataset, adjust the settings and start training. You can train supported LLMs, diffusion models, TTS models and embedding models without writing code. On supported LLM workloads, training is up to 2× faster and uses up to 70% less VRAM.
Export your trained models to NVFP4, GGUF and other supported formats. You can also securely deploy and access models remotely: turn on Remote access to publish Unsloth through a Cloudflare HTTPS link, then use the app and its local APIs from another device.
Hardware + platform support
Unsloth Desktop runs on Windows, macOS and Linux. Hardware support spans CPU and multi-GPU systems, NVIDIA and AMD GPUs, Intel hardware, and Mac.
CPU support includes Chat and Data Recipes. Training and inference options vary by model and backend.