Changelog

• macOS 26 design refresh and icon fixes
• Support for organization configuration
• Ultra model: improved translation capability
• Thinking models: remove <thinking> tag content

• macOS Tahoe Icon
• Increase max tokens for all Claude models
• Partial fix for Safari whitespace issues

• Custom mode icons, choose a custom icon for your modes
• Audio input device picker in settings topbar
• Improved audio device awareness
• Name of default audio device now shown in settings
• Fixed Turkish language missing from list
• Fixed language sometimes being translated to English in some presets

• Parakeet Multi-language, support for 25 languages (V3)
• Privacy improvement for API requests
• You can now process .ogg and .opus files
• Mode update hints in small recording window
• All modes now show in small recording window mode selector
• Copy to clipboard & reveal in finder actions on right click of history items

• Pause music/video while recording now works perfectly for sources
• Fixed cases where leading space was incorrectly added (Discord, Obsidian etc)
• Active App description now included Super mode and App context enabled prompts
• Fixed prompt rendering issues for Super mode

• Improved messages mode prompt
• Fixes for Custom model creation

• GPT-5, 5 mini, and 5 nano now in the app
• Support for Applications in nested folders in Applications
• Fix for Claude 4 series models in custom connections
• Visual tweaks for mode list view

Changes

• Mark pause media option as experimental
• Disable double click to zoom on windows
• Support for Perplexity Comet, Sigma, and Kagi browsers
• Set minimum version to macOS 13.3

Fixes

• Fix British english usage with Deepgram models
• Many small design improvements and cleanups

Features

• Parakeet is now 2x faster for long recordings
• Support for very long recordings on all voice models
• Progress bar integrated for all voice models on long recordings
• Improved UI responsiveness, non-blocking processing for long recordings
• Improved speaker separation and timestamp accuracy
• Faster diarization (speaker separation)

Fixes

• Parakeet hallucination issues on short recordings fixed
• Parakeet single word recording fixes
• Report issue form response error fixed
• Fix for animations, drag and drop, and collapsing sections in mode list

• Safari now supported in App activation
• Fix for short (<5 sec) recordings with Parakeet
• Fix for creation of new Ollama custom models
• Fixes for crashes in speaker diarization and mode management
• Fix for download location with newer models
• Fix for model management delete of custom models

• Fix change mode hanging during recording

Features

• Brand new audio backend, handles any audio device
• Safer and faster handling of audio data
• Live recording device switching support
• Faster response when stopping recordings

Fixes

• Design tweaks to the new settings interface
• Hover + keyboard mode selection bug fix
• Don't show Parakeet to users on macOS 13 and below
• Custom models can now be deleted from model list button
• Fix delete mode alert panel revealing

• All Custom API models now properly showing in model list
• Now showing alert when at mode limit
• Custom models can now be deleted from model list button

Features

• Beautiful new settings interface design
• Modes design overhaul - simple, sleek and more intuitive
• AI models are now managed directly from modes UI (AI Models tab removed)
• NVIDIA Parakeet (English) model now available for Pro users
• Super mode prompt and context improvements
• Ultra (Cloud) model, lower latency for users outside of North America

Fixes

• Various audio system crash safety improvements
• New prompting format strategy, better output with small models
• Fixes for transcribe from file functionality, UI improvements

• New design for settings panel
• Syncing now available between Macs for modes, models and vocabulary
• Faster active application icon updates in Super mode

• Custom models creation fix
• Filtering invalid segments fix

• Custom OpenAI compatible providers
• Result window pop in fix
• Crash fixes
• Error logging and performance tracking additions (opt in only)

• Mode ordering, drag and drop in list
• Fix Transcribe from file for .m4a files

• Vocabulary interaction tweaks and fixes
• Add noise floor for dynamic normalization in wave
• Mini recorder snap point fixes for multi-monitor
• Mini recorder shortcut hint flicker fix
• Mini recorder click outside of window to dismiss
• Handle crash due to model access after system Sleep

New Features

• New Recording window Mini mode
• Language support for English (British) & Chinese (Traditional)
• Menubar input device
• Dynamic audio normalization
• Add Claude series 4 models

Improvements & Fixes

• Reduced memory consumption during recording
• Faster recording session start times
• Fix Nova 3 vocabulary for non-english languages

• Offline voice models now load in background when recording starts
• Fix Nova 3 voice vocabulary usage

• Fix Nova 3 voice vocabulary usage

• Add Nova 3 voice model
• Fix app folder move logic
• Fix double space issue with Ultra (Cloud)

• Add GPT 4.1 series of models
• Minor UI clean-up
• Encrypt custom model API keys on save

• Add Dutch language back
• Fix handling for push to talk shortcut if microphone is slow to start
• Only mute audio if we cannot get now playing information from OS

• Crash fix on paste
• Fix super mode tag parsing
• Add experimental Traditional Chinese language option for whisper models
• Improve license blur style

• Ensure floating window is in view when presented, otherwise center it.

Improvements

• Animate menubar icon
• Animate app activation interactions

Fixes

• Fix app activation removal

• Crash fix

Features & improvements

• Can now appear in Full-screen apps
• Spaces are now added before pasted text if cursor is at end of a sentence
• If you are continuing a sentence, capitalization will be augmented properly

Fixes

• Fix for crashes on macOS 13
• Significant improvements to stability, addressing memory leaks, safety, thread safety, and specific crash scenarios

• Mode App activation settings improvement

• Clarity on which models are capable of translation

Features & improvements

• Massive speed improvements to for local models
• Claude 3.7 Sonnet added, Llama 3.2 added
• Improve offline language model stability
• Show window immediately, display initializing state
• Remove common hallucinations & segments outside of recording
• GPT-4o Transcribe models, now available in custom models

Fixes

• Crash fix for macOS 13
• Crash fix for audio file save

Features & improvements

• Massive speed improvements to for local models
• Claude 3.7 Sonnet added, Llama 3.2 added
• Improve offline language model stability
• Show window immediately, display initializing state
• Remove common hallucinations & segments outside of recording

Fixes

• Crash fix for macOS 13
• Crash fix for audio file save

Features & improvements

• Hold shift to auto-send
• Empty result sound effect
• Improve offline language model stability
• Added Llama 3.2 3b local model (experimental)

Fixes

• Fix streaming partial tokens issue
• Crash fixes and memory management fixes
• Fix '-' and ',' issue with Standard and Pro offline models

• Fix crash bug experienced by some users on onboarding launch

• History will update with recordings in realtime
• Fix mute when using bluetooth headphones, wait for output mode
• Fix app activation exceptions (WhatsApp, iTerm)
• Saved indicator now visible in mode configuration when on scrolled
• Improve Voice mode description
• Fix Super mode settings configuration- can now change all visible
• Fix issue with completed state of menubar icon when finished recording

• Report issue with recording from History (right click on sidebar)
• Screen will no longer lock automatically during recording
• If screen is locked, recording will continue after unlock
• Media will now pause when recording is started and play when recording ends
• Text Replacement duplication issue fixed
• Super mode will now respect your spoken language if not English
• Custom mode system prompt update

• Completely rebuilt onboarding experience
• New dedicated modes for Voice, Messages, Email, Notes, and Meetings
• Super mode officially released
• Claude 3.5 Haiku support
• New modes list design

• New implementation of the sound effect player

• Fix sound effect crash
• Improve crash-logs and reporting

• Fix for auto paste

• Fix for volume on license activation

• New super fast, 'Ultra (Cloud)' model
• New super fast, 'Ultra V3 Turbo' local model
• New medically focused 'Nova Medical (Cloud)' model
• New sound effects, you can switch themes in Sound settings
• New License UI in settings

• New 'About' & 'Changelog' view
• Fix check for updates window issue
• Allow defocus of the history view search input
• Better segment timestamps
• Fix copy to clipboard on segment view
• Add language model processing time to history sidebar

• Fixes for keyboard shortcuts interaction with other applications (BTT, Hyperkey, Karabiner etc)
• Fix for duplicate paste when using local language models
• Stability improvements for silence removal
• Update app icon to use dark variant

• Settings window design overhaul
• Global mode selection shortcuts (CMD + Number)
• New animated settings cards with improved styling
• Better window animations and transitions
• Improved sidebar navigation and layout
• Enhanced Push-to-Talk and Advanced settings animations
• Improved Whisper prompt string glossary formatting

• Animated recording window with fade effects
• Blue check indicator on mode selection

• Reduced pop-in for recording window bottom bar text
• Fixed sound playback issues
• Improved mode change behavior when no recording is active (immediately hide window)
• Better handling of shortcuts and keybinds

• Hitting change mode shortcut multiple times now cycles through mode options (same as pressing ↓)
• Mode select no longer starts a new recording
• Dont show active application on change mode window if using Super mode
• Fix: When changing shortcuts ensure old shortcut keybind is de-registered

• Fix for when Push to Talk disabled and shortcut is same as Toggle Recording

• New keyboard shortcut system with improved reliability
• Push-to-talk support with configurable shortcuts
• Redesigned settings cards with animations
• Better active recording indicators
• Improved volume slider controls for sound effects
• More reliable input field focus detection
• Faster event handling and performance improvements

• Bottom bar buttons styling and animation
• Delay and animate progress percentage
• Performance improvements for shorter recordings
• Recording context leak fix
• Volume fade in/out fix for Airpods
• Show millisecond accuracy in processing time and recording length in history

• Use macOS native text rendering (no longer renders Markdown)
• Massive performance improvements for Recording Window waveform
• Re-organize the bottom bar of Recording Window (add status LED)
• Failed API calls will now retry
• Initialize error tracking when user turns it on in settings
• Improvements to silence removal

• Additional error tracking

• Fix for auto paste closing if paste not enabled
• Success modal for deeplink license activation
• Fix for 'auto' language setting when using Groq or OpenAI Whisper models

• Automatically close recording window if paste was successful
• Fix for offline speaker separation

• Fade in/out volume

• v1.40.1: Fix for realtime view

History & Speakers separation

• Offline speaker separation (diarization), now works with any model
• Naming speakers in recording
• Segment being played is now highlighted
• Faster recording History (fast to load, fast to navigate)
• Fixes for selection & file delete in History (Follow's Mac standard Finder selection pattern)

Performance & new Models

• Countless speed improvements, will make app snappier especially surrounding start and stop recording
• Groq custom models expanded, added whisper, Llama 3.2 variants and more
• Long recordings now work with Whisper & Groq APIs
• Whisper V3 Turbo Model (experimental)
• Silence removal (experimental, recommended for V3 Turbo model)

Recording window

• Mini-recording window (use icon in top right to toggle)
• Recording window now saves position & minification state across application restarts
• New style for recording window

Modes

• Moved mute audio option to mode level
• Moved application context to prompt edit view

Other

• Removed some of the older experimental voice models
• Removed realtime option for offline models (this is currently a WIP, stay tuned)

• Fix for crash during onboarding
• Fix for performance drop
• Set default audio engine back to old audio engine

• Potential fix for crash on language selection in onboarding
• Fix for performance drop
• Set default audio engine back to old audio engine
• Onboading rebuild
• Utility to download model if not installed prompt on recording start
• Make new audio engine the default engine
• Add New Deepgram languages
• Fix: Don't send empty messages to LLMs

• Fix for performance drop
• Set default audio engine back to old audio engine

• Onboading rebuild
• Utility to download model if not installed prompt on recording start
• Make new audio engine the default engine
• Add New Deepgram languages
• Dont send empty messages to LLMs

• Bulk delete recordings from history
• Faster history window load time
• Experimental "Turbo processing" feature

• Add new audio engine option for multi-channel input devices

• GPT-4o mini now available
• Major improvements to Supermode formatting resolution
• Add Billing link to License View
• Fix for rendering of results and timing of resizing within UI

• Rollback fix for redraw height recalculation

• Fix issue where text would partially not display in recording window

• Show clipboard context update in realtime
• Show Mode name in recording history metadata panel
• Fix clipboard context use in custom modes

• Super mode selection & clipboard context integration
• Clipboard selction indicator in recording view

• Super mode beta release
• Deepgram realtime speaker separation & system audio
• More information available on recordings in recording history
• Deprecate Ultra Turbo model

• Add Anthropic Claude 3.5 Sonnet to Superwhisper Pro models
• Add Anthropic Claude 3.5 Sonnet to Custom AI models
• Allow changing application directory to location where folder already exists

• Fix for Push to Talk activation
• Fix for cancellation button on recording window & enter key activation
• Fix for display of error messages from API responses

• Push to Talk option for main shortcut
• New Super mode (experimental still - see launch video)
• Use system preferred language in onboarding
• All language support for Nova 2 realtime
• Vocabulary support for all models (whisperkit/deepgram)
• Simplify onboarding steps

• Pro tier includes GPT-4o and Claude Sonnet
• Nova 2 Realtime mode
• Recordings longer than 30 minutes will now work with Nova 2 model
• Tables, code, and titles now render in recording window output
• Experimental Super mode
• Improved performance for response rendering in recording window

• Include GPT-4o to models available for OpenAI

• History entry updates live as recording is being processed
• Save mode when deleting an example from the prompt
• Major improvements to messages prompt template in create mode menu
• Fix crash if audio is not recordable (faulty / unsupported devices)
• Fix issue with special characters in text replacements
• Fix for Enter hotkey enabling/disabling

• Add support for OpenAI Whisper with your own API keys
• Brand new Text Replacements interface
• Multi-line Text Replacements
• Text replacements only replace whole word matches e.g "'Cat' will be replaced with 'Dog', but 'Caterpillar' will not become Dogerpillar"
• Fix issue with Japanese language results
• Help text for adding GPT-4 and Ollama model connections
• Recommend smaller model to Intel users

• Add support for Ollama, unlocks using all latest offline AI models (Llama 8b, Phi-3)
• Add support for Groq, use with your API key to use Llama 70b, or Mixtral
• Add "Mute while recording" options in audio settings
• Add "Mute while recording" options in audio settings
• Allow result text in window to scroll
• Make sure system audio recording is stopped when recording is completed

• Add Llama 3 8b via Groq (stupid fast)
• Add images for models
• UI performance improvements for models tab

• Fix bug causing crashes introduced in v1.33.1

• Revert progress bar changes

• Performance improvements for switching between recordings in History
• Performance improvements for History open
• Rename speakers tab to Segments
• Progress bar speed adapts to estimate time more closely
• Remove default system prompts, fixes multi-language use of LLMs

• Massive improvements to History playback and UI
• Playback from point in recording
• Display progress of processing
• Save wav file every 10 seconds to avoid losing long recordings
• Word level timestamps

• Vocabulary feature, add words that are unique to you
• Make Claude 3 Haiku available to Superwhisper Pro users
• Move clipboard feature to prompt edit view
• Improve visibility of language model formatting feature
• Improve mode presets' prompts
• Only include clipboard contents if copied 10 seconds before recording or during recording
• Fix scroll on text replacements view

• Add Ultra Cloud model to Voice models

• Activate mode based on active website (focused tab in frontmost browser)

• Add menubar items to main menu
• Show app in dock by default
• Enable settings shortcut when app is focused (CMD + Comma)
• Dont steal focus when changing mode
• Show correct date for recordings in history
• Fix API key handling for Custom API connections

• Add menubar items to main menu
• Show app in dock by default
• Enable settings shortcut when app is focused (CMD + Comma)
• Dont steal focus when changing mode
• Show correct date for recordings in history
• Fix API key handling for Custom API connections

• Experimental Turbo Voice models with next gen Realtime mode
• Faster, voice activated improved Realtime mode for all models
• Realtime view performance improvements
• Realtime scroll fixes

• Claude 3 Haiku added
• Test Custom API keys before saving
• Show API errors in recording window when using cloud models
• Fix for navigation shortcuts enable/disable
• Make Language model connections creation more obvious
• Move expriemental switch to bottom of list

• Speaker separation feature available for Nova 2 models
• Segmented transcripts in history for all models
• Global shortcut and navigation keys for mode switch
• Escape key to go back to recording from mode select
• If user has completed onboarding, do not continue prompting for accessiblity permission
• Deprecation warning and hiding of literal punctuation feature
• Fix for long recording cancellation alert

• Activate when using Applications option added to Modes

• Custom API Connections for OpenAI and Anthropic (more coming soon!)

• Fix input volme adjust error handle

• Automatic input volume adjustment option

• Fix: Stop screen recording session bug

• Deepgram Japanese Support
• Access Cloud AI (Deepgram, OpenAI, Anthropic) access while on 15 minutes free
• Unload local voice model after 1 minute of no use, time delay is configurable
• More options for time delay on restore clipboard
• Fix change shortcut reset bug

• QR Code for License Key

• Cloud voice model added (Deepgram Nova 2)
• Cloud language model added (Claude Instant 1.2)
• Transcribe from File
• Open File with superwhisper
• Reprocess recordings from history

• Prompt examples interface
• Fix for pasting into unfocused applications (fullscreen / Raycast)
• Audio settings moved to their own settings tab
• Stability improvements for local language models
• Enhancements to "Create Mode" templates

• Include clipboard context in LLM
• Change mode shortcut & updates
• Location of app folder customizable
• UI style improvements for the recording window
• No longer send empty results to LLM

• History moved to it's own window
• Settings panel re-org and simplification
• User can configure second delay for restore clipboard on paste
• Cloud models shown in language models view
• Experimental models hidden by default
• Explanations of difference between voice and language models

• When creating new mode from template language model will be turned on by default
• Order of language and voice models has been fixed
• Menubar icon click to start recording is now opt-in
• Menubar icon shortcut displayed and updates with configured shortcut
• View history option in menubar
• Menubar mode switch only shown if user has more than 1 mode
• Turn on language model issue fixed

• Cloud hosted language model for Pro users only

• LLM crash bugfix

• Automatic paste now works on all keyboard layouts

• Onboarding flow activation bug fix
• License activation UI improvements
• Fix for menu bar button click

• Disable AI result switch if no AI result
• Default close window automatically to false

• Minor UI Fixes

• 3 new experimental language models (v2, German and Nano)
• Performance improvements to UI and model running
• 5x faster model download speeds globally
• Many UI and UX improvements
• Fixed paste triggering twice

• Minor UI Fixes

• 3 new experimental language models (v2, German and Nano)
• Performance improvements to UI and model running
• 5x faster model download speeds globally
• Many UI and UX improvements
• Fixed paste triggering twice

• 3 new experimental language models (v2, German and Nano)
• Performance improvements to UI and model running
• 5x faster model download speeds globally
• Many UI and UX improvements
• Fixed paste triggering twice

• Fix for settings init when loading with no recordings in folder

• History and Recordings View completely rebuilt

• Mode templates
• Create Mode callout
• Pro tier usage bar counts up instead of down
• App icon improvements

• Fix crash if model is not downloaded and you are trying to start a recording
• Fix language model access crash cases
• Bring back default window menus, allows for minimize, hide & close shortcutsc

• Customizable shortcuts for Toggle and Cancel Recording actions
• Enabled cmd + v paste shortcut into prompt editor

• Handle deeplinks for Raycast Extension
• Only start a new recording when changing modes if you are currently in recording or complete state
• Placeholder for wave when in Ready state, avoid UI popping

• Fix CPU usage when not in use issue

• Realtime mode teleprompter UI
• Recording Window UI refactor, footer glitches resolved

• Fix for audio devices that dynamically change output format

• Fix bug causing app to crash with audio devices that don't require downsampling

• Fix bug causing app to crash if changing audio devices

• Show output from language model as it outputs

• Improved downsampling, should result in better transcription
• If language model enabled, dont flash the transcription result
• If recording window is off, reset menubar app icon after 4 seconds

• Completely rebuilt onboarding
• Fix Downloading of language model
• Fix Model name setting in meta.json
• Fix mode voice model setting update

• Fix cancel warning, should only fire if in recording state

• New Models tab in Settings (Download & Delete models)
• Sidebar hide button
• Deeper integration with macOS (dock icon, main menu bar) etc

• Improvements to text replacements UI
• Show app icon in dock when onboarding or settings is open
• Copy to clipboard even if auto paste is off

• Text replacements feature

• Allow recording from the system's audio.
• Minimum system version is now macOS 13.

• Mode switch from Recording window
• License key is now copyable with a right click
• Show Recording window while model is initializing
• Performance improvements for file reading
• Massive refactor of recording state handling
• Dont allow deletion of last Mode from UI
• Limit mode creation

• Fix trapping in Edit case
• Fix for other file types in modes folder

• Fix initialization of mode being edited if you only have one
• Fix for language switch within modes, ensure correct mode is loaded on mode change
• Fix Save Mode when selecting language

• New feature: "Modes", allows you to make many configurations of models, languages, prompts, and configurations
• You can switch between "Modes" from the menubar
• When language model enabled, result will be pasted once complete
• Language model length has been adjusted allowing for longer expansions on what has been said
• Language model results are now visible in recording history
• Restore clipboard option will restore your clipboard value 3 seconds after result is pasted.
• Experimental feature: Simulate output keystrokes instead of using copy/paste (only works for US QWERTY layout)
• Fix: Escape global hotkey now properly de-activates when Recording Window is not enabled
• App icon has been matched to macOS standards

• Bring Recording History into settings view, remove menubar option
• Initialize language model when recording is stopped
• Clean LLM output text before copying to clipboard
• Remove Whisper model prompt

• Disable arrow key navigation when download or prompt edit modals are open
• Enable recording window and disable auto close when LLM is enabled

• Set selected model on settings appear

• New settings design
• Edit LLM prompt
• Reveal License Key

• Experimental LLM post processing
• Delete recording from dictation history

• Add delay before stopping recording to avoid losing last word
• Fix recording window resizing and animations

• Fix sizing for short recordings

• Improved dynamic resizing of recording window around text
• Fix animation of footer bar buttons on resize

• New and improved recording UI
• Hyper key can now used for the global shortcut
• Onboarding flow improvements
• Experiments moved to AI models settings

• When English selected, Pro and Ultra models are half as big and twice as fast.
• Details about selected model now displayed below dropdown selector
• More fixes to silence trimming, ensure last word does not get removed

• Fix performance of Dictation history window
• Recording window now updates immediately after recording is done processing
• Silence trimming at end of recording now pads with one second of silence to ensure words are not cut off

• All 100 languages are now available on free tier
• Pro tier models can be used for 15 min on free tier
• Status bar text on recording window indicates model type, language and application state

• All 100 languages are now available on free tier
• Pro tier models can be used for 15 min on free tier
• Status bar text on recording window indicates model type, language and application state

• Fix for model downloads

• UI fixes for recording window resizing

• Option to turn off Automatic paste
• Option to automatically close recording window after processing complete
• Experimental realtime implmentation
• Improved audio playback UI/UX
• Fix stability bugs and hanging issues
• Improved implementation of error tracking
• Single .wav file output now (previously two)

• Make size of audio wave larger
• Improve license management screen
• Remove Fn from keyboard shortcuts

• Option to show or hide realtime feedback window

• Animate soundwave between recordings in history view
• Fixes for fatal crashes on transcription start

• Realtime audio soundwave visualization
• Fixes for crashes in recording history view

• Fix for GPU/Metal on Intel based macs

• Massive performance improvement, use GPU/Metal for processing
• Remove "Turbo mode" option, for now.

• Resolve performance regression
• No app icon in Dock

• Search recordings
• New recordings are displayed in history immediately
• App version in recording stats
• Hover hints on recording stats
• Scroll only follows arrow key selection
• Empty states for first load up
• Suppress unhandled keypress noise when using arrow key nav

• Audio file playback
• Audio wave visualization

• Fix keydown handler in Dictation history view

• Dictation history view (Beta)

• Save recordings and transcribed text to its own folder
• Display alert if model fails to initialize

• Fix loading state check

• Onboarding flow
• Don't clear icon state if new recording has started
• Improved icon for loading state

• Indicator for model loading state
• Protection against multiple starts during model load
• Enhanced error logging

• New Turbo mode introduced
• Option to translate any spoken language to English

• New .dmg installer
• Small bug fixes

• New accessibility installation dialog and confirmation flow.
• New sound effects when recording begins and ends.

• Recordings are now converted to mono if the source audio is in stereo.

• Recordings now stored within superwhisper directory in Documents, not synced to iCloud.
• Error reporting now manually handled in more error capture locations.

• Automatic language detection option.
• Literal punctuation option.
• Automatically switch to system default input device.
• New settings layout and UI improvements.

• Added support for 100+ different spoken languages.
• Added "Nano" model, between the smallest "Fast" model and the "Standard" model.
• Fixed model download overlay not disabling the items behind the blur.