Real technical leadership. Meyer says XTTS was "the highest quality voice generation model in open-source" at the time he wrote (Josh Meyer). A community from day one. The team carried over a community from the Mozilla days, and open-source alignment helped it recruit scarce researchers (Josh Meyer). Adoption. It hit "something like 'open-source-market fit'" (Josh Meyer).

Coqui: the most popular open-source voice AI, and still no product-market fit

Key takeaways

Spun out of Mozilla's Machine Learning Group in 2021; shut down on 3 January 2024.
In the founder's words: 'millions of downloads didn't translate to millions in revenue.'
His lessons: models commoditize, the product matters more than the model, and pick a first market that can buy today.

What it is

Coqui trained and released open foundation models for text-to-speech and voice cloning, with a creator web app and a developer API on top; its best-known model was XTTS (Josh Meyer). The code remains on GitHub (coqui-ai/TTS).

Timeline

2016: the work began as the Machine Learning Group at Mozilla (DeepSpeech, Common Voice, TTS) (founder's shutdown post, via Hacker News).
2021: spun out as Coqui; STT v1.0 and a model zoo (founder's shutdown post, via Hacker News).
Jan 2022: a Show HN, "Clone your voice and speak a foreign language", got 513 points and 112 comments (Hacker News).
2023: Coqui Studio web app and API went live, with first customers; XTTS released openly (founder's shutdown post, via Hacker News).
3 Jan 2024: "Coqui is shutting down" (GitHub discussion).
Jul 2026: Meyer published his January 2024 post-mortem unedited (Josh Meyer).

Why it worked (as open source)

Real technical leadership. Meyer says XTTS was "the highest quality voice generation model in open-source" at the time he wrote (Josh Meyer).
A community from day one. The team carried over a community from the Mozilla days, and open-source alignment helped it recruit scarce researchers (Josh Meyer).
Adoption. It hit "something like 'open-source-market fit'" (Josh Meyer).

What to learn

Downloads are not revenue. "Creating the most popular open-source voice AI wasn't enough to close a new funding round" (Josh Meyer).
"The model is not king. Your product is." Wrapping a model in an API is not a defensible business when cloud providers and open models compete on price (Josh Meyer).
Choose a first market that can buy now. Meyer calls chasing AAA game studios, slowed by legal and IP concerns, "probably my biggest GTM failure"; he would pick product-led growth with SMEs (Josh Meyer).
Small teams should fine-tune, not train foundation models from scratch: "Models change, data is forever" (Josh Meyer).

Risks and what didn't work

The stated reason: "we never hit product-market fit, compute and talent were expensive, and the current funding atmosphere is tough" (Josh Meyer).
While productizing its model, competitors shipped newer architectures and Coqui "fell behind on the state-of-the-art" (Josh Meyer).
Outside view: commenters on Hacker News pointed to monetization trouble and confusing licensing (Hacker News).
These are the founder's personal views; he notes his team, investors and customers may see it differently (Josh Meyer).

FAQ

Short answers drawn from the sections above.

What is Coqui?

Why did Coqui work?

Real technical leadership. Meyer says XTTS was "the highest quality voice generation model in open-source" at the time he wrote (Josh Meyer).
A community from day one. The team carried over a community from the Mozilla days, and open-source alignment helped it recruit scarce researchers (Josh Meyer).
Adoption. It hit "something like 'open-source-market fit'" (Josh Meyer).

What can indie builders learn from Coqui?

Downloads are not revenue. "Creating the most popular open-source voice AI wasn't enough to close a new funding round" (Josh Meyer).
"The model is not king. Your product is." Wrapping a model in an API is not a defensible business when cloud providers and open models compete on price (Josh Meyer).
Choose a first market that can buy now. Meyer calls chasing AAA game studios, slowed by legal and IP concerns, "probably my biggest GTM failure"; he would pick product-led growth with SMEs (Josh Meyer).
Small teams should fine-tune, not train foundation models from scratch: "Models change, data is forever" (Josh Meyer).

What didn't work for Coqui, and what are the risks?

The stated reason: "we never hit product-market fit, compute and talent were expensive, and the current funding atmosphere is tough" (Josh Meyer).
While productizing its model, competitors shipped newer architectures and Coqui "fell behind on the state-of-the-art" (Josh Meyer).
Outside view: commenters on Hacker News pointed to monetization trouble and confusing licensing (Hacker News).
These are the founder's personal views; he notes his team, investors and customers may see it differently (Josh Meyer).

Sources

Every link cited above, in order of first use.

1.Josh Meyer (jrmeyer.github.io)
2.coqui-ai/TTS (github.com)
3.founder's shutdown post, via Hacker News (news.ycombinator.com)
4.Hacker News (news.ycombinator.com)
5.GitHub discussion (github.com)