Chip version numbers usually sound like marketing. In Mobileye's case the EyeQ generations actually encode a real story: the steady consolidation of an entire autonomy compute stack onto fewer, more capable pieces of silicon. Reading the progression is a clean way to understand how the industry moves from "driver assistance" toward "the car drives."
Mobileye's filings lay out the ladder. The company has described EyeQ5 as coming in two forms, "EyeQ5 Mid and EyeQ5 High," with the High part aimed at more demanding autonomy workloads. Later filings move the front end to EyeQ6, and the company describes systems "based on six EyeQ5 High SoCs," with a next generation "powered by three EyeQ6 High SoCs or one EyeQ Ultra, our AV-on-Chip." The most recent 10-K, filed February 12, 2026, is on sec.gov, surfaced through EdgarBeast.
Read that sentence carefully and the trend is obvious. An autonomous prototype that once required six high-end chips is targeted to run on three of the next generation — or on a single Ultra. That is not just a speed bump; it is the difference between a trunk full of compute and a production-viable module. Cost, power draw, heat, and packaging all collapse as the chip count falls.
The generational split also maps to use case. The Mid and High variants of a generation let a carmaker buy in at the level it needs: a mainstream lane-centering feature does not require the same silicon as a hands-off highway pilot, which in turn is a tier below a true robotaxi brain. One architecture, scaled across price points, is how a supplier serves both a $30,000 commuter car and an autonomous shuttle.
So when you next see "EyeQ Ultra" or a generation number in a spec, translate it. The label is really a claim about how much of the autonomy stack — perception, fusion, planning — the company believes it can put on one chip at automotive cost and power. The version history is the clearest public timeline of that ambition.