Someone used Claude Code to remake the Top Gun canyon to see which frontier LLMs can outfly Maverick

Someone used Claude Code to remake the Top Gun canyon to see which frontier LLMs can outfly Maverick

Published Nov 18, 2026, 12:51 AM EST Simon is a Computer Science BSc graduate who has been writing about technology since 2014, and using Windows machines since 3.1. After working for an indie game studio and acting as the family's go-to technician for all computer issues, he found his passion for writing and decided to use his skill set to write about all things tech. Since beginning his writing career, he has written for many different publications such as WorldStart, Listverse, and MakeTechEasier. However, after finding his home at MakeUseOf in February 2019, he would eventually move on to its sister site, XDA, to bring the latest and greatest in Windows, Linux, and DIY electronics. Summary LLMs can fly jets impressively, but they still crash—definitely not safe to fly people overseas. In a canyon test, Astra lasted the longest; Fable and Gemini missed the first gate, and Opus refused to pilot a warplane. The API costs for each flight: Opus 5.5 $0.00, Astra $1.92, Carbon $2.27, Fable 5.2 $14.57. Would you trust an LLM with flying the plane you're on? Yeah, me neither. Modern-day LLMs can do a lot, but they need to go a lot further still before I allow one to fly me overseas. And what better proof of that than watching the most powerful LLMs in the world nosedive into the floor while behind the controls of a fighter jet? Watching LLMs fly Super Hornets is a good reminder that they can't do everything They still did an impressive job, though Over on the ClaudeAI subreddit, user abhi3188 showed off what they made. The last time we caught up with them, they had challenged frontier LLMs to a boat race, which Gemini 4 Argon won. This time, they were tested to see which one could fly the plane from Top Gun: Maverick during the canyon scene, a Boeing F/A-18 Super Hornet. The competitors were ChatGPT Astra, Google Gemini Carbon, Claude Fable 5.2, Claude Opus 5.5, and Maverick himself (labelled as his actor, Tom Cruise). Maverick would fly as he did in the movie, but the AI had to figure it out as they went. Here's how it was set up: "Each model writes its own autopilot code. 10 times a second, the code gets the jet's full state (56 raw numbers like height, speed, G-force, clearance ahead, the next gate, the height ceiling etc) and sets the controls. No flying tips. Each model got 10 test flights to fix its code, with feedback on what happened, then one shot at the full run." The results were pretty interesting. At first glance, Opus 5.5 looks like it did a poor job flying, but abhi3188 claims it's because it refused to control a machine of war. Fable 5.2 and Gemini Carbon missed the first gate and were eliminated, leaving Astra to hold its ground (air?) against Maverick for an impressive amount of time before misjudging the distance while flying under a bridge and hitting the floor. You know what, I think these benchmarks are a lot more fun than the charts and numbers AI companies give us every release. But there's one question remaining: how much did each flight cost in API usage? Well, Opus cost $0.00, Astra cost $1.92, Fable cost $14.57, and Carbon cost $2.27.

Original Source

Read the full article at Xda-developers →

KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.