Whisper from OpenAI transformed speech recognition. Before Whisper, speech-to-text was an expensive enterprise solution requiring extensive R&D capabilities or deep pockets for hefty licensing fees. Even then, the results were mediocre at best. Whisper was the first ASR model that developers could actually deploy that "just worked". The impact has been profound over the last three years, with the repository accumulating over 86,000 GitHub stars and thousands of developers building dictation apps, meeting note takers, and voice-first interfaces.
We recently worked with a medical provider who wanted to bring affordable scribe software to underserved communities. Traditional solutions like Dragon Medical were prohibitively expensive for small clinics, requiring custom quotes and enterprise contracts. Because of Whisper, they were able to build a solution that is both affordable and accessible, bringing modern medical tools to communities that had been left behind. This exemplifies how open sourcing did not just make it easier for developers to build but made these capabilities accessible to users who could never afford them before. New providers like AssemblyAI and Deepgram have entered the ASR space, building on the foundation Whisper created, while NVIDIA's Parakeet now tops the OpenASR leaderboard, transcribing 60 minutes of audio in less than a minute on local devices.
The transformative potential of AI, from bringing medical scribes to rural clinics to enabling new forms of human-computer interaction, can only be realized when these tools are freely available to all who wish to build with them. Whisper demonstrated that open sourcing foundational AI models does not just advance the field technically but expands who gets to benefit from these advances. That matters more than any benchmark.