XPeng's Open-Source AI Bet Meets a Cautious Wall Street

XPeng has thrown open the code behind its latest perception model, publishing both the program and the model weights for GroundingPI. The disclosure, made alongside academic partners, gives investors a rare look under the hood of the company's artificial intelligence work — and a reminder that laboratory wins do not automatically translate into showroom revenue.

The model was built with the University of Peking and the University of Hong Kong, and XPeng intends it to underpin intelligent driving, robotaxis and humanoid robots. That reach across several development fronts is strategically tidy, though the release stops well short of a commercial deployment.

Benchmarks Point One Way, the Market Another

GroundingPI carries roughly four billion parameters. To make its case, XPeng points to 34 public tests spanning eleven categories — a breadth that matters, since the company is not leaning on a single result. On those published benchmarks the model scored 73.68, ahead of OpenAI's GPT-6 Astra at 71.54.

The caveat is built into the exercise itself. A benchmark lead is tied to the specific testing methods used and cannot be mapped directly onto the performance of a finished vehicle or robot. XPeng also reports encouraging results from robotics simulations, where systems running GroundingPI beat comparison models while using only 50% of the demonstration data; the rival systems consumed 75%. That hints at more capable learning with less training material, but the work remains, by the company's own framing, at the scientific prototype stage.

Should investors sell immediately? Or is it worth buying XPeng?

From Code to Curb

Where the research meets the road is in Guangzhou. XPeng has secured approval for citywide demonstration rides with Level 4 autonomous robotaxis, with public access rolling out gradually by invitation. The permit opens the door to real-world testing and to stress-testing the business model — the yardstick that matters most, since the technology's quality will show up where operating conditions and commercial demands collide.

A second outlet is planned for the first half of 2027, when XPeng expects to begin deliveries of its IRON humanoid robot in overseas markets. Neither the robotaxi clearance nor the IRON timeline, however, constitutes proof that GroundingPI is ready for series production in those products.

Deliveries Do the Talking

On the operating side, the numbers are harder to argue with. October brought a monthly record of 50,237 vehicles, up 22% on the prior month and 20% year over year. Overseas shipments reached 13,326 units, a 166% jump compared with a year earlier. That kind of momentum can underwrite further development spending — provided the financial benefit of that spending becomes visible.

Not everyone is convinced the growth story carries the day. J.P. Morgan cut its price target on XPeng to US$10 from US$11.50 on October 26, keeping a "Neutral" rating. The brokerage cited weaker demand in China and higher costs among Chinese automakers — objections that carry weight as long as demand and expenses cap the economic payoff of rising deliveries.

XPeng at a turning point? This analysis reveals what investors need to know now.

The stock has been soft in the meantime, with a daily move of -2.4% recorded in one session and -2.8% in another, offering little in the way of positive momentum.

The Verdict Awaits the Numbers

XPeng has set November 16 as the date for its third-quarter 2026 results, due before the US market opens. That report will be the real test of whether research, robotaxi trials and record deliveries add up to a convincing financial picture. For now, the case is cautiously constructive: the technology push is real, the volume growth is real, and the burden of proof rests on the bottom line.

Ad

XPeng Stock: New Analysis - 6 November

Fresh XPeng information released. What's the impact for investors? Our latest independent report examines recent figures and market trends.

Read our updated XPeng analysis...