Analysis

Mercedes-Benz and Liquid AI plan on-device conversational models for MBUX cars in North America, with a first deployment targeted for the second half of 2026

Mercedes and Liquid AI plan on-device speech and reasoning models for MBUX cars in North America; first rollout targeted for H2 2026. What is confirmed.

By ·

Mercedes-Benz announced on April 23, 2026 a multi-year partnership with Liquid AI to run conversational AI models inside the car instead of in the cloud, for MBUX third- and fourth-generation models in North America. The companies describe speech recognition, language understanding and reasoning as processed on the vehicle, and target a first production deployment as early as the second half of 2026. That is a target. In the sources we read, we found no announcement that it has shipped, and "without cloud" applies to the on-device part only: Mercedes says the models complement its cloud-based ones.

The essentials

What is Mercedes actually putting in the car?

Mercedes-Benz says the work evolves the MBUX Virtual Assistant, bringing together voice control, vehicle functions and contextual understanding. Liquid AI, a US start-up focused on small, efficient foundation models, supplies the models. The pitch is natural, low-latency conversation that does not depend on a continuous data exchange with the cloud. Liquid AI's chief executive describes the models as built to run on hardware already inside the vehicle.

Which cars are concerned?

The announcement's footnote lists model years. For third-generation MBUX it names the E-Class and CLE from model year 2024, the C-Class (combustion) for 2025 and 2026, the GLC (combustion) for 2025 to 2027, and the GT 2-door and SL for 2026 and 2027. For fourth-generation MBUX it names, among others, the CLA, the 2027 C-Class and GLB, the 2027 GLC (electric), EQS Sedan, GLE, GLS and S-Class. The list is about eligibility for the partnership. It does not say which cars will get which feature, or when.

Does it work without the cloud?

For the covered functions, that is the stated design: speech is processed on the vehicle, which Mercedes says makes everyday behavior more consistent. But Mercedes also says the approach complements cloud-based large language models for broader needs, so the assistant is described as hybrid. Neither company says in the sources we read which requests stay local and which are sent to the cloud.

What did Liquid AI's CEO add later?

A September 18 analysis cites an episode of the Latent Space podcast on which Ramin Hasani reportedly said the model is "about 600 megabyte", could be updated over the air, and that the first deployment would cover generation 3 North American cars "this year". We have not checked that episode ourselves. Treat the size and timing as secondhand. The same analysis notes that a size on disk is not a parameter count.

What is not established?

Summary table

Summary table
PointWhat is stated as of November 5, 2026Source
AnnouncementApril 23, 2026, multi-year partnershipMercedes-Benz USA media
ScopeMBUX gen 3 and 4, North AmericaMercedes-Benz USA media
On-deviceSpeech and language recognition, understanding, reasoningMercedes-Benz, Automotive World
CloudOn-device models complement cloud LLMsMercedes-Benz, Liquid AI
TimelineFirst production deployment as early as H2 2026Mercedes-Benz, Liquid AI
Model size"About 600 megabyte", per a secondhand reportFourWeekMBA (unverified)

Frequently asked questions

Is Mercedes' on-device assistant already in cars?

In the sources we read, it is described as planned, with a first deployment targeted for the second half of 2026. We found no launch announcement.

Does the assistant run entirely without the cloud?

No. Mercedes says the on-device models complement its cloud-based LLMs, so the assistant is hybrid.

Is this coming to Europe?

The announcement covers North America only. It says nothing about other markets.

Method and limits

Sources as of November 5, 2026. We read Mercedes-Benz USA's announcement page and used Liquid AI's release, Mercedes-Benz Group's page and Automotive World for context. The podcast figures come from a secondary analysis and we did not listen to the episode. We tested no vehicle and verified no on-device behavior.

Sources

Spotted an error in this article? Report it · Our corrections policy

Rekha icon: a house drawn in a single stroke

From the same author Self-promotion

An Eulerian path, drawn with your finger

Rekha, the iPhone game by OrigoTech's author: connect every dot in a single stroke without tracing the same link twice, across 500 drawings that are all solvable.

The OrigoTech newsletter

A sourced weekly digest of AI tool news.

Read next