This release adds 5 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+13 more
Summary
AI summaryBroad release touches Training & General Fixes, Studio, Hub + Download Manager, and studio.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Adds Chat with Files for document Q&A in Studio. Adds Chat with Files for document Q&A in Studio. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Adds Audio chat support for Gemma 4 (wav, mp3, m4a, flac, webm). Adds Audio chat support for Gemma 4 (wav, mp3, m4a, flac, webm). Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Adds an in‑app Update llama.cpp button for backend upgrades. Adds an in‑app Update llama.cpp button for backend upgrades. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Adds Cloudflare HTTPS free tunnels for end‑to‑end encrypted studios. Adds Cloudflare HTTPS free tunnels for end‑to‑end encrypted studios. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Feature | Medium |
Adds Tensor‑Parallel support for llama.cpp, boosting throughput ~30 %. Adds Tensor‑Parallel support for llama.cpp, boosting throughput ~30 %. Source: llm_adapter@2026-06-12 Confidence: low |
— |
| Feature | Low |
Adds a new Hub page for browsing, downloading, and managing Hugging Face models and datasets. Adds a new Hub page for browsing, downloading, and managing Hugging Face models and datasets. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Detects locally installed models/datasets and displays them alongside downloaded assets in the Hub. Detects locally installed models/datasets and displays them alongside downloaded assets in the Hub. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Provides direct Run / New Chat actions for downloaded GGUF models via the Hub. Provides direct Run / New Chat actions for downloaded GGUF models via the Hub. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Improves local tool calling reliability, ordering, and supports GGUF vision models. Improves local tool calling reliability, ordering, and supports GGUF vision models. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Enhances OpenAI‑compatible and Anthropic‑compatible API behavior for Studio servers, including better errors, token usage, stop reasons, and Claude Code compatibility. Enhances OpenAI‑compatible and Anthropic‑compatible API behavior for Studio servers, including better errors, token usage, stop reasons, and Claude Code compatibility. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Adds bypass permissions and tool call permission controls (Approve, Always Approve, Deny). Adds bypass permissions and tool call permission controls (Approve, Always Approve, Deny). Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Makes MCP and Artifacts selectable in the UI. Makes MCP and Artifacts selectable in the UI. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Implements S3 dataset loading for Studio (completes #5951). Implements S3 dataset loading for Studio (completes #5951). Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Bundles Gemma 4 chat templates (E2B/E4B + larger) and auto‑applies them to unsloth/gemma-4-*-GGUF models. Bundles Gemma 4 chat templates (E2B/E4B + larger) and auto‑applies them to unsloth/gemma-4-*-GGUF models. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Feature | Low |
Adds Apple GPU temperature and power monitoring in the Studio GPU monitor (macOS). Adds Apple GPU temperature and power monitoring in the Studio GPU monitor (macOS). Source: granite4.1:30b@2026-06-12-audit Confidence: low |
— |
| Feature | Low |
Adds inline confirmation options (Allow/Always allow/Deny) for tool calls in Studio. Adds inline confirmation options (Allow/Always allow/Deny) for tool calls in Studio. Source: granite4.1:30b@2026-06-12-audit Confidence: low |
— |
| Dependency | Medium |
Bumps install.sh / install.ps1 to require unsloth >= 2026.6.3. Bumps install.sh / install.ps1 to require unsloth >= 2026.6.3. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Dependency | Low |
Adds missing RAG dependencies to the no‑torch Studio runtime requirements. Adds missing RAG dependencies to the no‑torch Studio runtime requirements. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Performance | Medium |
Enables tensor parallelism for GGUF models, increasing throughput by ~30 %. Enables tensor parallelism for GGUF models, increasing throughput by ~30 %. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Performance | Medium |
Tunes llama.cpp environment settings for data‑center GPUs. Tunes llama.cpp environment settings for data‑center GPUs. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Performance | Medium |
Caches MCP tool discovery to avoid re‑probing on every chat send. Caches MCP tool discovery to avoid re‑probing on every chat send. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Performance | Medium |
Serves DiffusionGemma with live in‑place denoising and honest statistics in Studio. Serves DiffusionGemma with live in‑place denoising and honest statistics in Studio. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Bugfix | Medium |
Fixes llama.cpp update banner offering a downgrade or sticking on mixed releases. Fixes llama.cpp update banner offering a downgrade or sticking on mixed releases. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Medium |
Fixes false RoPE 'default' warning and rope drift gate on transformers 5. Fixes false RoPE 'default' warning and rope drift gate on transformers 5. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Medium |
Fixes decoding subprocess output as UTF‑8 on Windows in save.py. Fixes decoding subprocess output as UTF‑8 on Windows in save.py. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Medium |
Fixes step‑count mismatch when sequence packing is enabled. Fixes step‑count mismatch when sequence packing is enabled. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Bugfix | Low |
Reduces tool‑call nudging issues by 50‑90 % without losing accuracy. Reduces tool‑call nudging issues by 50‑90 % without losing accuracy. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Bugfix | Low |
Drops the on‑disk freshness cache after a llama.cpp update to avoid stale data. Drops the on‑disk freshness cache after a llama.cpp update to avoid stale data. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Bugfix | Low |
Fixes layout issues in the Studio S3 dataset panel. Fixes layout issues in the Studio S3 dataset panel. Source: granite4.1:30b@2026-06-12-audit Confidence: high |
— |
| Bugfix | Low |
Prevents negative counter‑reset readings for Apple GPU power. Prevents negative counter‑reset readings for Apple GPU power. Source: granite4.1:30b@2026-06-12-audit Confidence: low |
— |
Full changelog
We've merged over 150 PRs this week so lots of new updates, a new model Hub and look!
DiffusionGemma + Gemma 4 MTP + Audio
- Run and train DiffusionGemma via Unsloth Studio.
- Gemma 4 MTP is here! Run Gemma 4 around 2x faster with MTP - MTP is auto enabled in Unsloth Studio.
- Audio chat is now supported for Gemma 4 (
wav,mp3,m4a,flac,webm).
Hub + Download Manager (Experimental)
- Added a new Hub page for browsing, downloading, and managing Hugging Face models and datasets.
- Unsloth can now detect models and datasets already on your machine and show them alongside downloaded assets.
- Downloaded GGUF models now have direct Run / New Chat actions.
Chat with Files / RAG (Experimental)
- Added Chat with Files in Studio, letting you ask questions over your own documents and knowledge bases.
- Supports hybrid search, citations, PDF previews, per-thread documents, and a built-in
search_knowledge_basetool.
New Update Button + Hardware Support
- Unsloth now uses fresh, up-to-date llama.cpp prebuilts across CUDA, ROCm, Windows, Linux, and macOS.
- Added an in-app Update llama.cpp button so users can update the local backend without reinstalling Studio.
- Improved Windows / WSL AMD support, Strix Halo ROCm support, Blackwell CUDA selection, and clearer installer messages.
Local Chat, Tools & API Compatibility
- Local tool calling is more reliable, with better ordering of tool cards, fewer duplicate tool loops, and support for tool use with GGUF vision models.
- Improved OpenAI-compatible API and Anthropic-compatible API behavior for local Studio servers, including better errors, token usage, stop reasons, and Claude Code compatibility.
Tool Calling, MCP, Encrypted Cloudflare Tunnels
- Bypass Permissions, Tool Call Permissions (Approve, Always Approve, Deny)
- 50% to 90% less tool call nudging issues without any accuracy loss
- MCP, Artifacts are now select-able
- Tensor parallelism is now enabled for GGUFs - get +30% throughput!
- Cloudflare HTTPS free tunnels is now added allowing for end to end encrypted studios!
Training & General Fixes
- Improved MLX support with better model labels, generation speed stats, and fixes for VLM training.
- Fixed several training and dataset edge cases, including non-writable Hugging Face caches and custom dataset mappings.
- Added many UI polish fixes across chat, menus, model picker, dark mode, import/export, and settings.
To update Unsloth or install a new Unsloth Studio, you must use:
macOS, Linux, WSL:
curl -fsSL https://unsloth.ai/install.sh | sh
Windows:
irm https://unsloth.ai/install.ps1 | iex
[!WARNING]
DO NOT USEunsloth studio updatesince packaging will not get the latest updates
What's Changed
- Studio: llama.cpp update banner redesign, About tab license info, UI polish by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6196
- Bump install.sh / install.ps1 pin to unsloth>=2026.6.3 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6212
- Expose runtime context length for hub models by @alkinun in https://github.com/unslothai/unsloth/pull/6154
- Studio: fix llama.cpp update banner offering a downgrade / sticking on mix releases by @oobabooga in https://github.com/unslothai/unsloth/pull/6219
- Fix kwarg spacing in training files to satisfy pre-commit by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6209
- Studio: reword the Cloudflare line when the public probe fails by @danielhanchen in https://github.com/unslothai/unsloth/pull/6217
- fix: deduplicate lemonade ROCm prebuilt selection log by @LeoBorcherding in https://github.com/unslothai/unsloth/pull/6021
- Stop false RoPE 'default' warning and fix rope drift gate on transformers 5 by @danielhanchen in https://github.com/unslothai/unsloth/pull/6223
- fix(studio): load run.py by path for editable installs by @jimdawdy-hub in https://github.com/unslothai/unsloth/pull/5909
- fix(studio): inherit llama_extra_args and honor --no-mmproj by @jimdawdy-hub in https://github.com/unslothai/unsloth/pull/5902
- fix(studio): adopt server-loaded model before chat auto-load by @jimdawdy-hub in https://github.com/unslothai/unsloth/pull/5900
- Fix stale sidebar regression test to match the gap-px markup by @danielhanchen in https://github.com/unslothai/unsloth/pull/6232
- Studio: gate the staged prebuilt runtime validation behind a flag (off by default) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6216
- Fix FastModel config passthrough for sequence classification by @alkinun in https://github.com/unslothai/unsloth/pull/6203
- fix: decode subprocess output as UTF-8 in save.py on Windows by @dylanschroers in https://github.com/unslothai/unsloth/pull/6218
- patch: fix EmptyLogits gathering in nested payloads and Accelerate recursively_apply by @MdHussain121 in https://github.com/unslothai/unsloth/pull/6092
- Studio: show Apple GPU temperature and power in the GPU monitor (macOS) by @Ban921 in https://github.com/unslothai/unsloth/pull/6187
- Studio: Add inline confirmation (Allow/Always allow/Deny) for tool calls by @oobabooga in https://github.com/unslothai/unsloth/pull/5869
- Studio: guard Apple GPU power against negative counter-reset readings by @danielhanchen in https://github.com/unslothai/unsloth/pull/6235
- Fix step count mismatch when sequence packing is enabled by @IrakliXYZ in https://github.com/unslothai/unsloth/pull/5967
- fix/uv-bytecode-timeout by @alkinun in https://github.com/unslothai/unsloth/pull/6166
- Studio: tune llama.cpp env for data-center GPUs by @danielhanchen in https://github.com/unslothai/unsloth/pull/6098
- Studio: drop the on-disk freshness cache after a llama.cpp update by @danielhanchen in https://github.com/unslothai/unsloth/pull/6234
- Add missing RAG deps to no-torch Studio runtime requirements by @danielhanchen in https://github.com/unslothai/unsloth/pull/6236
- Studio: rounded rectangle hover states for menu items instead of pills by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6210
- docs: repository cleanup by @Agnibha007 in https://github.com/unslothai/unsloth/pull/5617
- Run cross-platform parity test on Windows and macOS in CI by @danielhanchen in https://github.com/unslothai/unsloth/pull/6241
- chore(studio/frontend): normalize line endings to LF by @danielhanchen in https://github.com/unslothai/unsloth/pull/6012
- fix: respect absolute export paths to prevent cross-drive copy failures (WinError 112) by @anmolxlight in https://github.com/unslothai/unsloth/pull/6088
- Studio: Add Tensor-Parallel llama.cpp support by @oobabooga in https://github.com/unslothai/unsloth/pull/6040
- Studio: Add custom provider option to Connections by @Imagineer99 in https://github.com/unslothai/unsloth/pull/6112
- Studio: model selector and settings polish by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6240
- Studio: login card polish and sidebar label alignment by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6242
- Studio: pinnable plus menu items and saved prompt pins by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6237
- Studio: bottom update banners, smooth llama.cpp progress, re-prompt after copy by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6233
- fix(studio/responses): forward chat_template_kwargs enable_thinking to chat request by @Anai-Guo in https://github.com/unslothai/unsloth/pull/6202
- Studio: fix WSL Strix Halo GPU on reinstall (ROCDXG drop-in + system HIP before bundle) by @danielhanchen in https://github.com/unslothai/unsloth/pull/6227
- Studio: fully rounded Hub pills and refreshed menu icons by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6248
- Studio: use px-2.5 for Hub option menu padding by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6249
- Studio: fix Downloaded model list disappearing and order it by last download by @danielhanchen in https://github.com/unslothai/unsloth/pull/6247
- Studio: new-chat shortcut, composer draft autosave, archive threads by @NilayYadav in https://github.com/unslothai/unsloth/pull/5771
- Studio: persist speculative decoding preference across restart and model switch by @oobabooga in https://github.com/unslothai/unsloth/pull/6169
- Studio: refine menu chevron, tick icon, and one-line plus-menu shape by @shimmyshimmer in https://github.com/unslothai/unsloth/pull/6251
- Studio: serve DiffusionGemma with live in-place denoising and honest stats by @danielhanchen in https://github.com/unslothai/unsloth/pull/6250
- Studio: bundle Gemma 4 chat templates (E2B/E4B + larger) and auto-apply to unsloth/gemma-4-*-GGUF by @danielhanchen in https://github.com/unslothai/unsloth/pull/6245
- feat(studio): implement S3 dataset loading (completes #5951) by @ashzak in https://github.com/unslothai/unsloth/pull/6222
- Studio: cache MCP tool discovery instead of re-probing every chat send by @oobabooga in https://github.com/unslothai/unsloth/pull/5828
- Fix Studio S3 dataset panel layout by @wasimysaid in https://github.com/unslothai/unsloth/pull/6252
- Attach DiffusionGemma visual-server from the prebuilt bundle by @danielhanchen in https://github.com/unslothai/unsloth/pull/6254
New Contributors
- @jimdawdy-hub made their first contribution in https://github.com/unslothai/unsloth/pull/5909
- @dylanschroers made their first contribution in https://github.com/unslothai/unsloth/pull/6218
- @MdHussain121 made their first contribution in https://github.com/unslothai/unsloth/pull/6092
- @Ban921 made their first contribution in https://github.com/unslothai/unsloth/pull/6187
- @IrakliXYZ made their first contribution in https://github.com/unslothai/unsloth/pull/5967
- @Agnibha007 made their first contribution in https://github.com/unslothai/unsloth/pull/5617
Full Changelog: https://github.com/unslothai/unsloth/compare/v0.1.451-beta...v0.1.46-beta
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Unsloth
All releases →Related context
Earlier breaking changes
- v0.1.43-beta Do not use `unsloth studio update`; it does not fetch latest updates.
- v0.1.40-beta Auto MTP speculative decoding enabled by default for MTP GGUFs; warns on stale llama.cpp prebuilt
Beta — feedback welcome: [email protected]