Simba 3.2 is #1 on Artificial Analysis — the independent TTS leaderboard the whole industry watches. Not our benchmark. Not our numbers. Theirs.
And on Voice Arena — the blind, listener-voted one — we’re joint second. Which sounds like a smaller story until you look at who is up there with us. Take the ones you cannot actually ship, and the ones that cost 7x, off the board, and Simba 3.2 is the only real-time voice at the top. Number one on one benchmark. Effectively number one on the other. Number one on the row that matters: real-time, at price, at quality.
Two independent benchmarks. Same answer.
The reason this matters: nobody at Speechify picked the winner.
Artificial Analysis runs a consistent, objective TTS evaluation across the industry and updates the board as models ship. We are #1 on it today. If a new model tops us tomorrow, the board will say so — that is the contract.
Voice Arena ranks text-to-speech the honest way. Real native-speaker panels hear two clips generated from the same text, blind, without knowing which model made which, and pick the one that sounds more natural. Enough of those votes roll up into an Elo rating — the same method the Chatbot Arena made the standard for ranking large language models. It is stricter than most: six languages, a balanced voice slate per model rather than whatever default sounds best, sentences written for the places TTS actually ships, and a methodology built with advice from Prof. Shinji Watanabe at Carnegie Mellon.
No self-reported MOS. No cherry-picked sample. No “internal evals show.” Just objective evaluation, and real people listening.
Why “joint second” on Voice Arena is a first-place story
Look at who is sitting where.
The model tied with us for second costs 7x what we charge. The one model rated above us does not run in real time, so if you are building anything that talks back, a voice agent, a live reader, a call, it is not even on your shortlist.
Take the models you cannot actually ship out of the running, and count only what is left: the real-time voices a team can put in production today. On that list, at this quality, nothing gets near our price. Simba 3.2 is from $6 per 1M characters, the best-sounding real-time voice you can put in front of users, at a price built for production volume rather than a demo.
Best-sounding real-time and lowest-cost are not supposed to be the same row on the table. This time they are. See the full breakdown on the pricing page.
We’ll move you over
If you are shipping on another provider today, the gap between “that sounds better and costs less” and “we actually switched” is real work: voices to re-map, SSML to port, latency to re-check under your own load. So we close it for you.
- Drop-in tools. A clean REST API , first-party TypeScript and Python SDKs , and the Speechify Cookbook, runnable migration recipes with SDK and native-REST versions side by side, so you can see exactly what goes on the wire.
- Forward-deployed engineers. For teams with volume, our FDEs embed with you and own the migration alongside your team: voice mapping, prosody parity, load testing, cutover, until it holds quality at scale.
You do not have to choose between the best voice, the best price, and a painless switch. Bring the use case, we will do the lifting.
Hear it yourself
Do not take our word for it. That is the whole point of an independent leaderboard. Check Artificial Analysis, vote on Voice Arena, then come try Simba 3.2 with a free API key. Compare it against whatever you run today, on your own text.
We think you will hear number one. Apparently, so does everyone else.