AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Block has a page at block.xyz/dessa presenting Dessa as “the most authentic full-duplex speech model ever built.” The supplied source material contains a privacy notice, not technical documentation or evidence for that claim, so the model’s capabilities, availability and evaluation remain unconfirmed.

Block’s Dessa page describes a product as “the most authentic full-duplex speech model ever built,” but the material provided for this report does not include a model announcement, technical specification or test results to support that description. The available text is a privacy notice for dessa.com, leaving the model’s capabilities and release status unverified.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

The phrase “full-duplex speech model” appears in the supplied headline, alongside Block’s authenticity claim. The source text does not explain what the model does, how it handles simultaneous speaking and listening, or how Block defines “authentic.” No benchmark scores, demonstrations, research paper or independent assessment are included.

The document that is available identifies Block, Inc. and its affiliates as the parties responsible for information processing associated with the Dessa website. It says the notice applies to use of that website and not to other Block services such as Square, Cash App, Afterpay or TIDAL, which have separate privacy notices.

According to the notice, website information may include details users provide, such as contact information, as well as device and usage data collected automatically. Block says it may use this information to operate and improve the site, communicate with users, analyze usage, and protect its services. Those disclosures describe the website’s data practices; they do not establish how a speech model is trained or what audio data it processes.

At a glance
announcementWhen: The source privacy notice lists an effe…
The developmentBlock’s Dessa page promotes a full-duplex speech model with a superlative authenticity claim, but the supplied material does not substantiate it with technical details or results.

What the Claim Could Mean

If Dessa is a speech model designed for full-duplex interaction, its performance could matter for conversational applications where people expect to interrupt, respond naturally and speak while a system is listening. That use case differs from systems that wait for a user to finish before generating a reply. However, the supplied source does not confirm that Dessa supports those behaviors or describe any product built around it.

The superlative claim also raises a question of how authenticity is measured. Natural-sounding speech, realistic turn-taking and responsiveness are distinct qualities, and the source provides no evaluation method or comparison set. Readers and developers cannot assess the claim from the privacy notice alone.

For now, the concrete news is limited: Block has a Dessa-branded page whose headline makes a broad performance claim. Without access details or evidence, prospective users cannot determine whether a model is available, what it can do, or how it compares with other speech systems.

Products Worth Considering

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

Bose QuietComfort Wireless Noise Cancelling Headphones - Twilight Blue Limited Edition

Bose QuietComfort Wireless Noise Cancelling Headphones – Twilight Blue Limited Edition

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, and IPX4 Water Resistance

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, and IPX4 Water Resistance

As an affiliate, we earn on qualifying purchases.

What the Dessa Page Says

The supplied source is a page at block.xyz/dessa, but the report text consists largely of a privacy notice rather than a product description. The notice says it covers information collected through the Dessa website and that continuing to interact with the site constitutes consent to the practices described. It lists an effective date of October 21, 2026.

Block’s notice describes collecting information directly from users and through their use of the site, including device characteristics, IP address, browser details and visit activity. It also says information may be shared with Block affiliates and service providers for stated business purposes. These are website disclosures, and they should not be read as technical documentation for the model.

No model card, launch date, developer access information, pricing, audio-processing policy or named research team appears in the material supplied. The source therefore supports reporting that the page makes the claim, but not that the claim has been independently demonstrated.

““The most authentic full-duplex speech model ever built.””

— Dessa page headline

Evidence Behind the Authenticity Claim

The supplied material does not establish whether Dessa is publicly available, when it was announced, or whether the page headline is an official product launch claim. It also provides no model architecture, supported languages, latency figures, audio samples, benchmark results or technical explanation of full-duplex operation.

It remains unclear what Block means by “authentic,” what systems or human judgments the claim is based on, and whether any external researchers have evaluated the model. The privacy notice does not say whether speech inputs are collected, retained or used for training. Those questions cannot be answered from the supplied source and should not be inferred from its general website data disclosures.

Details Needed to Evaluate Dessa

A fuller product announcement or technical release from Block would be needed to confirm availability and intended use, explain how the model handles live conversation, and disclose relevant audio-data practices. Evidence such as reproducible evaluations, comparison criteria and independently reviewable demonstrations would help readers judge the authenticity claim.

The supplied source does not state when such details will appear or identify a planned launch, paper or public test. Until Block provides that information, the model’s status and performance remain unverified beyond the wording on the Dessa page.

Key Questions

What is Dessa?

The page headline describes Dessa as a full-duplex speech model. The supplied source does not provide a technical description or explain its intended applications.

Is Dessa available to use?

The provided material does not say whether the model is available, who can access it or when a release might happen.

Has the “most authentic” claim been verified?

No supporting benchmark, methodology or independent evaluation is included in the source material. The phrase should be treated as a claim on the page, not as an independently confirmed result.

Does the privacy notice explain how Dessa handles voice recordings?

No. It describes information collected through the Dessa website, including contact, device and usage information, but does not specify whether the model processes or retains audio.

Source: hn

You May Also Like

The Truth About Europe’s Frontier Lab And Its AI Ambitions

An analysis of Europe’s leading AI lab, Mistral, reveals it lags behind global competitors in AI development, raising questions about European sovereignty in AI.

Why Nvidia’s Flora Makes Robot Video Labeling 125× Cheaper, With A License Catch

Nvidia says Flora can label robot video for 72 cents an hour, versus $88.49 for GPT-6 Astra. The cost is modeled, and the license bars commercial use.

Local AI On The M5 Ultra Mac Studio: What 512GB Actually Buys You

An in-depth look at what the 512GB configuration of the M5 Ultra Mac Studio offers for local AI models, including capabilities and limitations.

Could you spot an AI-written book?

Recent tests reveal even close readers struggle to distinguish AI-generated text from human writing, raising questions about AI’s role in authorship.