Eleven v4 vs Sesame Preview: This Isn't the Comparison You Think It Is

Searching for ElevenLabs Eleven v4 against Sesame Preview turns up plenty of pages that will happily put them in a table. Almost all of them are comparing a product to a benchmark against a demo, and the table is the least useful thing on the page. Eleven v4 is a general-availability speech synthesis model with an API, a rate card and a published arena rank. Sesame Preview is a free consumer voice assistant that answers when you tap a link. They are not two options for the same purchase, and the honest version of this article is about which one you actually want — because the answer depends on a question neither vendor's marketing page asks.

What Eleven v4 actually is

ElevenLabs shipped Eleven v4 and Eleven v4 Turbo on September 28 as production endpoints. Eleven v4 carries 90-plus languages, a 10,000-character cap per request, multi-speaker dialogue with stable speaker identity, inline delivery directives, improved IPA phoneme support and instant voice cloning from ten seconds of audio. It ranks first on Artificial Analysis' Provider Voice Arena at Elo 1,319 with 1,674 samples behind the estimate, and second on the Controlled Voice board at 1,157. On the API rate card it lists at $0.022 per thousand characters during a promotion that ends October 12, reverting to $0.08 — the same rate the previous ElevenLabs Eleven v3 has carried.

Every part of that paragraph is checkable. There is a model identifier, a documented input limit, a published per-character price, a deprecation note for the older Turbo tiers, and an independent leaderboard entry with a confidence interval. That is what a model looks like when you can buy it.

What Sesame Preview actually is

Sesame Preview is a consumer demonstration. It is a voice assistant you reach through the company's own site, with a small cast of named personas — Maya, Miles, Simone and Charlie — conversational memory that carries between sessions, web search, notes and reminders, and a summary after the call. It is free. Signed-in sessions run about half an hour; guest sessions about five minutes. It handles English only. And Sesame states plainly that it does not execute tasks: the point of the demo is the conversational texture, not an agent that books your meeting.

The part that matters for this comparison is what is missing. There is no model identifier, no endpoint, no rate card, no input limit you can design against, and no announced date for the production model. Sesame's published research line is a separate thing — an open-weights conversational speech model with a 4K context — and that is not the same system you talk to in the Preview. So the thing named "Sesame Preview" on every comparison table is a product you cannot integrate with, and the thing you can download is not the product.

Why they cannot share a scoreboard

The controlled-voice arena on Artificial Analysis is the closest thing this field has to a neutral measurement, and Sesame Preview is not on it. It is not on the provider-voice board either. There is no Elo, no vote count, no price normalisation — and no way to produce one, because a leaderboard needs a callable endpoint and a per-unit price and Sesame Preview has neither.

So any table placing these two side by side is filling the Sesame column from a product tour. Comparing a $80-per-million-character endpoint against a free web demo on "naturalness" is not a comparison; it is a category error with an Elo-shaped hole in it. If a page claims a winner in that matchup, it invented the basis for the claim.

What can be said honestly is what each one is trying to be. Eleven v4 is optimised to turn a script into a performance, at scale, through an API, with the same voice every time. Sesame Preview is optimised to make a stranger feel like they are talking to someone, once, in a browser tab.

The one place the comparison is real

There is a genuine question underneath the search, and it is about open weights. If what you actually want is a speech model you can download and run yourself, neither vendor's flagship is the answer — ElevenLabs' stack is closed, and the Sesame system you can run is the research model with a 4K context, not the assistant.

For that job the useful reference point is on the arena boards: Breeze TTS 2 is the highest-ranked open-weights entry on Provider Voice, at Elo 1,206 and $34.00 per million characters, with 16 open-weights models in that board's field of 92. That is 113 points behind Eleven v4 and well ahead of the Qwen3 TTS line at 944 and 930. If your project needs to run on your own hardware, the comparison worth making is Eleven v4 against Breeze TTS 2 — not against a browser demo — and the tradeoff there is real: you give up roughly a hundred Elo points and the audio-tag workflow, and you gain control of the whole stack.

Choosing, given what each one is for

• If you are building a product that speaks — Eleven v4, on an API, on a rate card. Nothing about Sesame Preview is integrable, so it is not in this decision.

• If you want a conversational demo to show someone what voice AI feels like — Sesame Preview, and it is free, which is the entire argument for it.

• If you are evaluating emotional range before committing to a vendor — Eleven v4's Elo 1,319 on the vendor-voice board and 1,157 on the cloned-voice board are the numbers to read, with the caveat that the second is the one that describes a cloned-brand-voice product.

• If you need speech in languages beyond English — Eleven v4 documents 90-plus; Sesame Preview is English-only and says so.

• If you need the model to actually do something — neither, in this matchup. Sesame Preview deliberately does not execute tasks, and Eleven v4 is a synthesis engine that needs your own stack around it.

• If you need to run it locally — neither. Breeze TTS 2 is the open-weights reference point.

• If cost is the deciding factor — Eleven v4 is $0.022 per thousand characters until October 12 and $0.08 after, and Sesame Preview is free because it is not a product you can buy. A free demo is not a cheaper option; it is a different category.

The one thing both share is a date problem. Eleven v4's promotional pricing expires on October 12, which changes its cost by a factor of nearly four. Sesame has not given a date at all for the production model, which is a different kind of uncertainty — the Preview could be superseded next month or sit unchanged for a year, and there is no way to plan around a launch nobody has scheduled.

If your stack ends up spanning more than one voice vendor

OrcaRouter hosts neither ElevenLabs nor Sesame, so there is nothing in this article to call through us and we will not imply otherwise. What we do cover is the layer above a stack that is already multi-vendor. One API key reaches 200-plus models with provider list prices passed through at 0% markup, so a price change upstream — this launch's promotional window being a live example — lands on our side the same day rather than at the next invoice. Where the speech path is customer-facing, automatic failover shifts traffic to a healthy upstream when one degrades, which is how you can audition a new endpoint on real traffic without betting a release on it.

What to take away

Eleven v4 is a model you can buy, measure and depend on, and it currently sits at the top of the vendor-voice board with a two-week price cut. Sesame Preview is a free experiment in conversational feel, English-only, capped at thirty minutes, and not something you can build on. The reason they appear in the same search results is that both are voice AI and both are new — not that either is an alternative to the other.

If you came here to choose between them, the useful reframe is this: you are not choosing a voice model. You are deciding whether you are shipping a product or trying a demo, and only one of those decisions has a rate card attached. If it is the product, read the Eleven v4 numbers and re-read them on October 13, when the promotional rate expires and the comparison against every other vendor on the board changes.