Willow

Willow is a voice dictation app for Mac, Windows, and iPhone that lets users speak into any text field, from Gmail and Slack to Cursor and ChatGPT, in place of typing. This launch introduces Atlas 1, the company's new in-house speech-to-text model that now powers every Willow user by default. The pitch is aimed at people who live in written communication all day (founders, developers, clinicians, students) and want dictation accurate enough to trust on technical vocabulary, accented speech, and noisy environments. According to the launch, Atlas 1 posts a 1.2% word error rate on clean audio and 2.1% in real-world conditions with background noise, compared to the 5 to 15% range Willow attributes to competing systems like Whisper, Deepgram, ElevenLabs, and AssemblyAI on the OpenASR benchmark. Willow credits much of the accuracy gain, particularly on Asian and Asian-American accents, to a distributed human operations layer it describes as the first scalable human-powered transcription infrastructure built for real-time dictation. A full benchmark report is published on the company site. Willow is built by Allan Guo and Lawrence Liu , Stanford dropouts who initially focused on healthcare software for assisted living facilities before spending a year exploring ideas and landing on voice dictation . The company is backed by Y Combinator and markets itself to enterprise buyers with HIPAA and SOC 2 compliance, so Atlas 1 matters now because it moves Willow off third-party transcription models and onto infrastructure it controls, which is the piece most voice products outsource.