XPeng's CEO Tried Tesla's FSD in Paris. Here's His Verdict

XPeng and Tesla were the only two brands in the Paris Motor Show's official Autonomous Lab, the public-road assisted-driving programme that ran both companies' systems on real Paris streets. He Xiaopeng was there — and he did something unusual for a chief executive at his own product's launch event. He rode in a Tesla with FSD engaged, and then said so publicly, with a video.

His verdict, posted on 13 October: "I also tried Tesla FSD — overall the performance was very smooth. XPeng NGP and Tesla FSD have both made breakthrough progress in technical architecture, and each has its strengths. XPeng's model is larger and has stronger generalisation ability."

He closed with the industry framing: good technology improves through competition, and everyone should push global intelligent driving to a higher level together.

▶ Play the four-minute video — filmed partly from inside the Tesla with FSD running on Paris streets.

Inside the Tesla: the centre screen's driving visualisation with FSD active on a Paris street. Image: He Xiaopeng on Weibo.

What is actually in the video

It is a produced piece with bilingual Chinese and English subtitles, and it cuts between two settings: a drive-along inside a Tesla on ordinary Paris streets — parked cars, a cyclist, side roads, the car's own visualisation on the centre screen — and He Xiaopeng standing outside the venue talking, wearing an XPeng stand lanyard marked "Physical AI", at one point saying "you should try it" to the person he is with.

The subtitle track catches fragments of his technical commentary rather than a structured review. One frame carries "双 Orin 大概提高了三倍 / that of the previous dual Orin" — a compute comparison, with the English subtitle reading "that of the previous dual Orin". Another reads "to achieve easier global generalization", and another "at the architectural level".

So the video is evidence of the ride, not a test. There are no intervention counts, no route, no timings, and no statement of how long he spent or whether he was driving or a passenger.

How to read a CEO's verdict

Three things are worth keeping in view, because this is the most interested possible observer.

The compliment is real but cheap, and that is the point. "Very smooth" is the word everyone reaches for after a short ride in a modern assisted-driving car — it is the same word XPeng's own curated reviewers used, and the same word the company uses in its campaign copy. A chief executive saying a competitor's system is smooth at a show where both are demonstrating costs him nothing and buys the appearance of fair-mindedness.

The claim is the specific part. "XPeng's model is larger and has stronger generalisation ability" is not diplomacy; it is the company's core technical argument, and it is the same argument its engineers made to Bloomberg the same week — that NGP can be adapted to a new European country in about two weeks, using roughly one ten-thousandth of the training data of the underlying model (we covered that here). The CEO's verdict and the engineering claim are the same claim.

The venue is the signal. Tesla is the benchmark in this category, and XPeng is the challenger. A challenger's CEO publicly comparing his system to the incumbent's, in the incumbent's own car, on the same streets, is a confident move — it invites a comparison he would lose on brand and could only win on capability. He came away saying both are good and his is bigger.

What the week established

This sits alongside the third-party verdicts from the same programme, which were compiled and circulated by XPeng-side accounts and included a North American Tesla commentator saying NGP "easily rivals V14, and in some places is even better". We have also been here before: in Amsterdam, a Xinhua-organised comparison with an XPeng engineer driving both cars produced a similar "genuinely close" reading, with the same structural caveat that the test was run by one side (our coverage is here).

Taken together, three things are being said at once and only one of them is measurable: XPeng and Tesla are close; XPeng's architecture is more general; and the comparison should not be taken too far, by anyone, because nobody has run the controlled test that would settle it.

Caveats

• He Xiaopeng is the most interested party in this comparison. He is describing his own product against his own benchmark, in a video published by his own account.
• A four-minute ride is not a test. No route, duration, intervention count or metric is given, and the post does not say whether he drove or was a passenger.
• The video is a produced piece with bilingual subtitles, cut together from a drive and interviews; it is not continuous footage of a drive.
• "Very smooth" is subjective and is the standard phrase for a short assisted-driving ride, in both praise and marketing.
• Both systems are Level 2. FSD's availability in Europe remains limited and market-by-market, so the comparison is of pre-release capability in one city, not of what customers can buy.
• The "larger model" claim is not quantified — no parameter counts, compute figures or benchmark results accompany it.
• XPeng's own demonstration is not independent either. The Autonomous Lab public test rides, like the Amsterdam comparison, were organised with the participating companies.

The bottom line

The interesting part of this is not the verdict — "both are good, ours is bigger" is what any confident challenger would say. It is that a Chinese chief executive chose to say it from the passenger seat of a Tesla, on camera, in Paris, while his own system was taking the public around the same streets. Two years ago that comparison was the thing being avoided. Now it is the marketing, and the measurable version of it is still missing.

Source: He Xiaopeng's Weibo post and video of 13 October 2026, with our earlier coverage of the Paris Autonomous Lab verdicts, the Amsterdam comparison and XPeng's European NGP timeline — translated & adapted for XPENGTech.