Developer

Dessa AI API Pricing: Has Block Set a Rate for Developers?

2026-11-12 · 7 min read · By DessaAI Daily

Developers searching for Dessa API pricing are asking a practical budgeting question: if Block opens its speech-only voice model to builders, what will a minute of conversation cost? The honest answer is short, and this article will not dress it up. Block has announced no Dessa API at all — so there is no API price, no rate card, no per-minute charge and no billing model to quote. What follows is what that absence means, how the voice APIs that do exist price themselves, and how to budget sensibly without inventing a number Block has never published.

First, the status: no API, therefore no API price

Start from the record. Dessa was announced on October 23, 2026 as a research preview: a full-duplex, speech-to-speech model trained from scratch on human conversation audio only, with no text and no pretrained models. In our Dessa AI API status guide we set out what Block has and has not announced, and that position has not changed. There is no public API, no developer documentation, no SDK and no API key programme. Our later guide to Dessa API documentation, SDKs and keys found the same answer for each artifact a developer would normally read pricing from: none of them exist yet.

That matters because API pricing is never a free-standing number. It sits inside a product: which model you can call, what counts as a billable unit, whether listening and speaking are charged differently, and what limits apply. Until Block announces that product, any specific Dessa API price circulating online is speculation or a fake, whatever confidence it is written with.

How existing voice APIs present their prices

The contrast is useful, because the APIs developers can actually use today do the opposite of staying silent: they publish rates. OpenAI's realtime voice offering, Hume's EVI and ElevenLabs' conversational products are all available now, and each presents a published way of charging — the details differ, but the pattern is the same: a defined unit, a listed price for that unit, and terms a finance team can model before a line of production code ships.

We compared those published approaches in voice AI pricing compared, using the providers' own published rate information at the time of writing, and in Dessa AI API vs OpenAI Realtime API. The point for budgeting is not to copy a rival's number onto Dessa. It is to notice what a real pricing announcement looks like — an official page, a unit, a rate and terms — and to wait for Block to publish the same kind of page before treating any Dessa figure as real.

Why local inference complicates the pricing guess

There is a second reason not to guess. Block says Dessa's inference runs faster than real time locally on a MacBook Pro with an M4 Max chip or better, after training on a single NVIDIA GB200 NVL72 system. A model that can run on a customer's own hardware does not have to be sold as a hosted, per-minute API at all. Block could, in principle, choose several different routes — hosted access, licensed local use, a product inside its existing businesses, or continued research access — and each route would produce a completely different price structure, or no public price at all.

None of those routes has been announced. Local inference is a demonstrated capability described by Block, not a shipping promise, and it does not tell us whether a developer API is planned, let alone what it would cost. Our general guide to what Dessa costs reaches the same conclusion for consumers: no pricing of any kind has been announced, and joining the waitlist is free.

How to budget while you wait

If you are planning a voice product now, budget from things that exist. Price your prototype on a published rival rate, because that is money you would actually spend today. Keep the integration layer replaceable, so a future API — from Block or anyone else — can be evaluated on its real published terms rather than on hope. And treat any Dessa line in a budget as unpriced and unscheduled: no rate, no date, no commitment you can defend to a manager or a customer.

That discipline is also the scam check. Block has announced no paid developer tier, no early-access API pass and no reseller programme. If a site offers to sell you Dessa API credits, keys or a discounted per-minute bundle today, it is selling something Block has not released. The only official place for Dessa is Block's page at block.xyz/dessa, and early access is going first to a small group of AI safety researchers.

What would change this answer

Three announcements would change it, in order. First, Block announces developer access to Dessa at all. Second, it publishes documentation and terms defining what developers can call and how usage is counted. Third, it publishes a rate. Until all three exist, the correct entry in any pricing table for a Dessa API is the one this blog uses everywhere else: not announced. When Block does publish a rate, we will update this article the same day with the official figure, its unit and its source — and not before.

Sources: Block's official Dessa page (block.xyz/dessa) for the research-preview status, local inference and training claims, checked November 2026; DessaAI Daily's published API status, documentation and pricing-comparison articles, which cite the providers' published rates at the time of writing.

Frequently asked questions

← More Dessa AI articles Join the official waitlist ↗