//opengauntlet
// surveyed, not benchmarked — 168 systems

Now make it speak

The leaderboard scores how human a model reads. It says nothing about how a model sounds — that is a different piece of software entirely, and choosing it badly is where most voice projects go wrong. Here is every text-to-speech system that matters, on the axes that actually decide a project.

Surveyed, not benchmarked. Nothing on this page was measured by OpenGauntlet.

How to read this section

None of this is benchmarked here. OpenGauntlet measures Conversational Language Humanlikeness — transcript-level qualities like word choice, warmth and emotional reasoning. It deliberately does not score prosody, intonation or timing. This section is a sourced survey of the speech market, not a trial. Where a claim could not be verified against a primary source, it says so rather than smoothing it over.

How to read these numbers → · What you're allowed to do with them →

//

Every system, compared

Filter by kind or by what a system can actually do, pick the columns you care about, search any field, click a heading to sort. Every row carries the one thing a buyer would otherwise find out too late.
Loading…
commercial-safe weights permissive, with a catch not usable commercially proprietary service discontinued
//

Claims this corrects

Each of these is repeated widely, and each is wrong.
//

What we couldn't establish

A comparison that hides its gaps is less useful than one that names them.