Rewards program

We pay for what makes local models better

If you improved a local model for recognition, diarization or translation beyond any available open-source, we pay from $100 to $10,000 — depending on the result.

How much we pay

$100 – $1,000

Noticeable improvement

Speed-up, lower memory use or an accuracy gain on our scenarios.

$1,000 – $5,000

Major leap

The model clearly beats the best open-source on quality, or works where hardware couldn’t cope before.

$5,000 – $10,000

Breakthrough

You rewrote the approach and got a result many times better than any open solution. Essentially top-tier software we partly buy out.

Why we do this

Our goal is a fully local speech-processing pipeline on the user’s device. The main obstacle is the quality of local models on ordinary hardware. Breakthrough improvements come from specific people with specific ideas, so it’s better for us to honestly pay for what actually moves the product forward.

What we evaluate

The main criterion is a reproducible result on real tasks: speech recognition in calls, speaker separation, real-time translation. We look at quality, speed and resource use on typical consumer hardware, not benchmarks in a vacuum.

How to take part

  1. 1

    Message us on Telegram with what you improved and share your GitHub

  2. 2

    We invite you to a private repo in our organization — put the code there with steps to reproduce and measure

  3. 3

    We verify reproducibility and agree on the reward based on the result

Tell us about your result
Bounties for better local speech models — up to $10,000 — SpotMax