Superwhisper

• Hundreds of thousands use Superwhisper to transcribe messages, meetings, AI prompts and more every day across iOS and macOS.
• Superwhisper achieved 2x increase in their Free Trial to Pro conversion rate after upgrading their default onboarding to use Parakeet powered by Argmax.
• Argmax has built a differentiated proprietary implementation of the open-source Parakeet model, which brings unique advantages to Argmax customers.

‍

What the Founder Has to Say

‍

Argmax has built the best inference pipeline for Parakeet available on the market. It gives our users efficient, high accuracy, voice to text on their machine with privacy they can trust.

-Neil Chudleigh, founder of Superwhisper

‍

‍

‍

From Cost-Saving Solution to Profit Center

‍

In 2024, Argmax set out to upgrade on-device inference from a cost-saving solution to a profit center for the ASR industry by migrating the best-in-market cloud workloads to the edge. When Argmax matches cloud API accuracy but delivers superior latency on device, user retention skyrockets. The Superwhisper case study is the first in a series of studies we will publish over the next few months with real-world user growth and retention data.

Superwhisper enables users to switch between several cloud and on-device inference services with a single keystoke under a single flat monthly subscription fee. Hence, Superwhisper users freely shop around and only stick with an inference service if it consistently provides the best experience. This is a perfect experimental setup that removes cost as a factor and focuses only on the user experience and retention.

From Month 1 to Month 6, Argmax's MAU in Superwhisper grew 567% as tens of thousands of users organically discovered the new inference service. With the recent Argmax upgrade as the default onboarding service, the growth accelerated in January 2026.

The onboarding of Free Trial users with a paid inference service like Argmax initially presented a cost challenge to Superwhisper. However, the 2x increase in conversion rates to Pro led to a net profit while accelerating user growth.

‍

How does Argmax differentiate inference for open-source models?

‍

Most inference platforms onboard dozens, if not hundreds, of models every year. In sharp conrast, Argmax has onboarded a whopping 4 models since our inception 2 years ago. Here's why:

Our insight is that each high-potential open-source model makes for a great launch demo. However, commercial adoption at scale requires the following:

• Identifying and fixing the model's inherent limitations discovered in the wild under real-world conditions, ultimately leading to a commercial-grade implementation.
• The model's feature set, accuracy and speed being superior to everything else on the market at the time of launch such that a significant part of the market can justify the upgrade.

When Nvidia released Parakeet v2 in May 2025, it ranked 1st on the OpenASR leaderboard. 9 months later, Parakeet v2's accuracy is still among the top, only marginally surpassed by a handful of 5-15x larger and slower models.

At Argmax, we immediately recognized Parakeet's high potential based on our in-house evaluations on customer private test sets. Within a few weeks, we shipped our first implementation in alpha and identified the following limitations based on customer feedback:

• Short-form dictation inputs (<3 words) frequently led to 0 output words.
• Real-time streaming mode yielded a higher word error rate compared to pre-recorded audio.
• The lack of custom vocabulary (keywords) support broke feature parity with Whisper and cloud ASR platforms, blocking lift-and-shift migration.

We fixed these showstopper issues in 3 months. The results:

• Dramatically reduced the incidence rate for "0 output words".
• Developed a real-time streaming algorithm to achieve accuracy parity across pre-recorded and real-time streaming modes. [Demo]
• Developed a keyword recognition algorithm on top of Parakeet that supports up to 1000 keywords which is 10x the number supported by Whisper and other cloud ASR platforms. [Blog]

‍

‍

‍

Find out more about Argmax SDK today.

‍

‍