Step 3.5 Flash officially launches on a large scale today, and Zeekr 8X Super Eva starts mass production and is now available for sale

ME News Report, April 17 (UTC+8), April 17, 2026, Zeekr's all-new flagship SUV Zeekr 8X officially enters mass production and delivery. Equipped with multiple cutting-edge models such as Step 3.5 Flash, the full-vehicle intelligent agent Super Eva has officially completed the critical leap from technical release to large-scale mass production, and is now being delivered to consumers. This marks a milestone for Step 3.5 Flash, which after topping the global OpenRouter weekly call volume chart in March this year through open source, has transitioned from the open source community to large-scale commercial deployment in end-user vehicles. The Zeekr 8X equipped with Super Eva is also China’s first mass-produced vehicle to deliver the "Grok+FSD" experience. Over the past year, the domestic automotive AI scene has seen considerable attention on large models, but transitioning from laboratory demos to consumer-ready mass production has always been a barrier for the industry. The official launch of Super Eva today signifies that the vehicle’s intelligent agent, powered by Step, now has the capacity for large-scale deployment in real user and real driving scenarios. Super Eva was jointly developed by Step, Geely, and Qianli Technology, and is the first full vehicle intelligent agent integrating smart cockpit, assisted driving, and digital ecosystem, natively fused with underlying control systems such as intelligent driving, chassis, and power. Its technical concept aligns closely with Tesla’s Grok conversational control architecture, earning it the nickname "Grok+FSD" in China, and it is first launched on the Zeekr 8X. Super Eva is equipped with three core models from Step. The Step 3.5 Flash inference model is designed specifically for agent scenarios, with a maximum inference speed of 350 TPS per request, capable of accurately understanding vague commands and completing multi-step task planning in real time. For example, a user saying "Take me to pick up the kids from school, stop by McDonald's, and arrive at school before 5 PM" can trigger full autonomous execution of route planning, waypoints, and assisted driving activation. Step’s speech large model breaks free from the traditional three-stage architecture of ASR-LLM-TTS, delivering lower latency and more natural emotional expression, truly achieving "understanding tone and having personality." The visual understanding large model endows Super Eva with perception capabilities of the interior and exterior environment, from judging road conditions to recognizing parking spaces, making decisions more comprehensive and reliable. It is understood that Super Eva’s capabilities will continue to be upgraded and integrated with more ecological services. It will no longer be just a companion inside the car but a seamless connection hub to the world, proactively understanding scenario-based needs such as booking restaurants, ordering takeout, reserving flights, and buying tickets, all through natural dialogue during driving, making everything effortless. Industry experts believe that 2026 will be a critical year for the "entry" of general large models. The mass production of Super Eva validates Step 3.5 Flash and other models’ stable deployment capabilities in high real-time industrial terminal scenarios, providing a replicable path for Step’s deeper penetration into more hardware scenarios. In the future, Super Eva’s capabilities will expand infinitely, not only as an in-car AI partner but also as an intelligent hub connecting the physical world. It will integrate with Geely’s proprietary after-sales and car rental ecosystems (Geely Ecosystem) and fully connect lifestyle services (Digital Ecosystem), eliminating the need to switch apps or go through cumbersome booking processes. On the road, users only need natural dialogue to have Super Eva proactively understand scenario-based commands, handling everything from restaurant reservations, takeout, flight bookings, to ticket purchases. (Source: Ifnar)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 12
  • Repost
  • Share
Comment
Add a comment
Add a comment
ByteBard
· 12h ago
Zeekr 8X's delivery speed is impressive; how long has it been from open source to getting on board with Step 3.5 Flash?
View OriginalReply0
GateUser-276116e2
· 19h ago
The name Super Eva sounds sci-fi; what is the actual voice interaction latency in milliseconds?
View OriginalReply0
YieldGoblin
· 05-22 09:50
What is the actual experience of Grok+FSD? Can any car owners share their real-world road condition performance?
View OriginalReply0
GateUser-9187acf1
· 05-22 07:11
From being the top in weekly call volume to mass production and vehicle deployment, this conversion efficiency is unthinkable for traditional automakers.
View OriginalReply0
GateUser-88d5d071
· 05-22 07:05
The homogeneity of in-vehicle large models is serious. Where is Zeekr's differentiation? Has anyone disassembled it?
View OriginalReply0
GateUser-318a7dc8
· 05-22 07:04
Has the Zeekr 8X pricing been announced? Will the costs of intelligent driving hardware be passed on to the vehicle price?
View OriginalReply0
PuddingMarketMaker
· 05-22 07:03
In-vehicle AI is finally no longer just PowerPoint presentations; the journey from laboratory to delivery has taken too long.
View OriginalReply0
UnderTheWisteriaBridge
· 05-22 07:01
The concept of the full vehicle intelligent agent is quite new, but the features users perceive most strongly are still navigation and parking.
View OriginalReply0
PixelUniverseCat
· 05-22 07:01
China's first mass-produced Grok+FSD, Zeekr is aiming to seize the dominant voice in autonomous driving.
View OriginalReply0
Miner'sHelmetUnderTheMoonlight
· 05-22 07:00
April 17, 2026? I'm confused by this timeline—was it a typo or a preview?
View OriginalReply0
View More
  • Pinned