The worm in your dashboard

Mercedes-Benz says its next MBUX in-car systems will understand your voice command using a model that lives inside the dashboard — recognising the speech, parsing the intent, and reasoning about it locally, with no round-trip to a datacenter. The partnership was announced in April, with first production in North America targeted for the second half of 2026. Shopify is running a related model that finishes a commerce search in under 20 milliseconds. The vendor behind both is Liquid AI, and in a Cognitive Revolution interview its CEO Ramin Hasani used the two deployments to argue that a specific bet has left the demo stage.

A 230-million-parameter model that Liquid says out-scores a Google model several times its size on tool-use — on Liquid's own benchmarks, not yet independently re-run.

The bet is what Hasani calls 'device-native': models designed from the architecture up to live on a phone, a car, or a wearable — not big cloud models compressed down to fit. Almost everyone is now chasing on-device AI (Apple, Google, Microsoft, Qwen, IBM all ship small models), so the differentiator isn't the goal but the lineage. Liquid's models descend from an MIT and TU Wien effort to mathematically copy the nervous system of a worm, C. elegans; an early demo steered and parked a car using a control network of just twelve artificial neurons. The commercial models are larger than twelve neurons, but the same instinct — do more with radically fewer moving parts — is the pitch.

The number Hasani leans on is a 230-million-parameter model, small enough to sit in a phone's memory, that Liquid reports out-scoring a Google model several times its size on tool-use tasks. Those benchmarks are Liquid's own and haven't been independently re-run, so treat them as a claim rather than a settled result. What is checkable is the shipping traction: a major carmaker and a major commerce platform have both committed to running these models on their own hardware. If intelligence has spent five years migrating into the cloud, the question this raises is whether the roughly $800 billion of phones and laptops people already own is where a lot of it quietly moves back.