HyperWhisper Linux 1.0.0

HyperWhisper's first public Linux release for Ubuntu 22.04+ and Debian 12+ on x86_64/amd64.

Install

Download hyperwhisper_1.0.0_amd64.deb, then run:

sha256sum --check SHA256SUMS
sudo apt install ./hyperwhisper_1.0.0_amd64.deb
sudo usermod -aG hyperwhisper-input "$USER"

Log out and back in after joining hyperwhisper-input. That group is optional, but global shortcuts and automatic paste require it.

See the Linux installation guide for GNOME/KDE companions, permissions, privacy details, and removal.

Verification status

The exact release commit passed the hosted Rust, .NET, integration, Avalonia UI, production coverage, virtual PipeWire microphone, Debian package, and Lintian gates in GitHub Actions run 32685798268.

The expanded physical GNOME, KDE, Vulkan, and CUDA hardware matrix is intentionally deferred to v1.1.0 and tracked in issue #273. Please report v1.0.0 hardware-specific defects on GitHub.

The static APT repository archive is not attached because this build did not use an APT signing key. Install the .deb directly; no public APT endpoint is being advertised for this release.

SHA-256 (hyperwhisper_1.0.0_amd64.deb): baeda1c7ca3c7f2ad94d410ba6c7d2c021b080b0e5af2a57cae6bb7b5c1c0054

Windows 1.11.0

Highlights

• Custom vocabulary reaches more providers: Grok, Mistral (your own key), Google Chirp 3 and OpenAI gpt-transcribe now receive your terms. Previously they were built and dropped.
• Three new post-processing models: Gemini 3.7 Flash, Grok 4.6 and GPT-5.6 Luna. Five retired models redirect to a current one automatically.

Features & Improvements

• The Provider picker names the company and the Model picker names the model. Google appears once instead of twice, and the Model row is always a dropdown.
• New modes start with the English spelling for your system region. Existing modes are untouched.
• Links in the update window are now clickable.

Bug Fixes

• A terminal provider error during streaming no longer triggers a doomed reconnect whose generic failure message replaced the real one.
• Genuine licence-validation faults are now reported, so a backend outage is visible instead of silent. Ordinary licence verdicts stay unreported.

macOS 2.44.1

Highlights

• Faster AssemblyAI transcription: short clips now return on AssemblyAI's fast path instead of falling back to the slower upload-and-poll route on every request

Bug Fixes

• Fixed a Parakeet issue that pinned one CPU core and made the menu bar menu unclickable while a Parakeet mode was selected
• Fixed the History window pushing its content off-screen — selecting a transcript clipped the sidebar, the search bar and the detail header

macOS 2.44.0

Highlights

• Custom vocabulary reaches more providers: Grok, Mistral (your own key), Google Chirp 3 and OpenAI gpt-transcribe now receive your terms. Previously they were built and dropped.
• Three new post-processing models: Gemini 3.7 Flash, Grok 4.6 and GPT-5.6 Luna. Five retired models redirect to a current one automatically.

Features & Improvements

• The Provider picker names the company and the Model picker names the model. Google appears once instead of twice, and the Model row is always a dropdown.
• New modes start with the English spelling for your system region. Existing modes are untouched.
• Links in the release-notes window are now clickable, and the speed-data row is a single line.

Bug Fixes

• HyperWhisper Cloud fails fast with a clear message when no licence is active, instead of recording, converting and uploading first.
• A rejected or expired licence key now reads as an ordinary result rather than a server fault.
• A terminal provider error during streaming no longer triggers a doomed reconnect whose generic failure replaced the real message.
• The audio file picker no longer blocks the app while you browse.
• History metadata badges wrap onto a second line in a narrow window.

Windows 1.10.0

Added

• Find the fastest provider for you. The new latency page shows how quickly every cloud provider actually answers — by server region and clip length, measured from real transcriptions instead of vendor benchmarks. Use it to pick the fastest provider for where you are.

  Your transcriptions help build it. Share anonymous speed data (Settings → General) is on by default and stores one timing per provider call: no account, no key, no audio, no text. Turn it off any time — see Data & Privacy.

Fixed

• Release notes render bold text instead of <b> tags (#174)

macOS 2.43.0

Added

• Find the fastest provider for you. The new latency page shows how quickly every cloud provider actually answers — by server region and clip length, measured from real transcriptions instead of vendor benchmarks. Use it to pick the fastest provider for where you are.

  Your transcriptions help build it. Share anonymous speed data (Settings → General) is on by default and stores one timing per provider call: no account, no key, no audio, no text. Turn it off any time — see Data & Privacy.

Fixed

• Global hotkeys register without the main window (#182)
• Launch-minimized no longer hides a window you just opened (#184)
• Apple Speech no longer reports your own cancellation as a provider failure (#177)
• Release notes render bold text instead of <b> tags (#174)
• The launch-time recording scan moved off the main thread (#171)

Windows 1.9.0

Highlights

• August 2026 model refresh: OpenAI gpt-transcribe and gpt-live-transcribe, AssemblyAI Universal-3.5 Pro, Claude Sonnet 5, Gemini 3.6 Flash, Gemini 3.5 Flash Lite, Grok 4.5, Mistral Medium 3.5 and Qwen 3.6 27B.
• Deepgram live dictation fixed: streaming always timed out after 10 seconds, and SpeechStarted / UtteranceEnd / endpointing frames failed to parse, dropping words mid-sentence.

Features & Improvements

• Remove filler words in live dictation: the setting previously applied only to non-streaming transcription.
• More reliable streaming shutdown: wider WebSocket close-code handling, a safety net so a stopped recording always reaches a terminal status, and a fixed startup race when recording immediately after launch.

Bug Fixes

• Long Groq dictations no longer fall back to the raw transcript: Groq applies an undocumented 2,048 completion-token default, so long transcripts came back finish_reason=length and were rejected with Response rejected (OutputLimit); keeping original transcription. Requests now send an explicit cap.
• Completed transcriptions can no longer be overwritten: a failure writer could replace an already-saved Completed result.
• Vocabulary replacements: ApplyReplacements now trims only when vocabulary is actually configured.
• Paragraph breaks: the local API /post-process endpoint dropped dictated paragraph breaks.
• Backup export: fixed a crash (NullReferenceException) in backup export settings.
• Reduced Sentry noise from benign no-speech results.

macOS 2.42.0

Highlights

• Redesigned onboarding: A new 8-step first-run walkthrough.
• August 2026 model refresh: OpenAI gpt-transcribe and gpt-live-transcribe, AssemblyAI Universal-3.5 Pro, Claude Sonnet 5, Gemini 3.6 Flash, Gemini 3.5 Flash Lite, Grok 4.5, Mistral Medium 3.5 and Qwen 3.6 27B.

Features & Improvements

• Punctuation & formatting control: Choose how much punctuation and automatic formatting you want in dictation.
• Faster short clips: AssemblyAI now uses a fast path for clips under 120 seconds.
• Full localisation: All 39 languages are complete on macOS.

Bug Fixes

• Fixed the app freezing during live transcription.
• Vocabulary replacements now apply when AI post-processing is off.
• "Remove filler words" now applies to live transcription.
• Fixed History search rows showing another recording's content.
• Fixed license validation at launch, and a rejected replacement key no longer removes a valid cached license.
• Missing Parakeet models now stay local and ask you to download them.
• Fixed Deepgram live transcription parse failures.
• Fixed dictated paragraph breaks being dropped by the local API.
• Fixed stale cancel confirmation for recordings.

Windows 1.8.3

Highlights

• Faster AssemblyAI transcription: clips under two minutes take a single-request fast path instead of upload-and-poll.

Bug Fixes

• Vocabulary replacements with AI off: modes with post-processing disabled discarded every configured replacement.
• Groq Llama 4 Maverick removed: the model was decommissioned upstream, so post-processing silently returned the raw transcript. Existing modes migrate to GPT-OSS 120B.
• History no longer destroys completed transcriptions: a finished result could be overwritten with Transcription did not finish and its text wiped.
• Mouse-wheel scrolling on Modes and Model Library: both pages could previously only be scrolled by dragging the scrollbar.

Windows 1.8.2

Bug Fixes

• Cancelling a transcription no longer reports a failure. Pressing Escape mid-transcription, or Cancel during a file transcription, could show Transcription failed: Cannot dispose while processing, please use DisposeAsync instead. and mark the recording as Failed in History. Cancelling now stops cleanly and the recording is left as cancelled.