In Bijan Bowen’s YouTube video review comparing GPT-6.1 Sol with Claude Sonnet 5.5, Sonnet trounced Sol overall across four tasks, to the creator’s surprise. For work that needed doing right now, they would choose Sonnet.
Bowen found Sonnet’s robot branding and marketing package much more refined and its music and video significantly better. The demolition-derby games were a closer contest, and a remote-control-car parking challenge had no clear winner. Sol was still a good, useful workhorse in Bowen’s view.
Bowen said the models were now at the same price, although Sonnet might use three times as many tokens and cost more to use. They ran both on $200-a-month subscriptions and wanted to compare what the models could make, not how many tokens they used.
Sonnet’s derby game came closer to the image
Bowen asked each model to turn a generated demolition-derby image into a game. The instructions called for cars that resembled the image, multiple camera views, crash damage and sound, suspension and visible engine parts.
Sol produced a working game with crash effects, an exposed engine, suspension and parts that fell off. Bowen thought the result was impressive as a game but a poor match for the reference image. Sol’s response acknowledged using approximations and falling short of the exact-match request.
Sonnet’s cars looked more faithful to Bowen, and its cinematic camera shots added to the result. Bowen said, “I have to say this is pretty fun to play”. They were impressed by what both models had built, even as they judged Sonnet’s version the closer replication.
Detached parts in Sonnet’s game could still block the cars after falling away. Bowen noted the bug but continued to find the game fun and its cars closer to the image.
Bowen would put Perry’s website live
For the robot-product task, Bowen supplied files and instructions for a desktop chatbot project. Each model was asked to make new software, branding, a website and a marketing video. Sonnet named its chatbot Perry.
Bowen explored Perry’s interactive website and found its branding and product presentation exceptionally strong. They said they personally could not have made a marketing site that looked as good and would put this one live if launching the product. Bowen’s verdict on the design was emphatic: “from a brand design standpoint here, knocked it out of the park. Home run grand slam”.
Perry’s software worked in Bowen’s live test: they changed its volume and set a timer by voice. The praise extended to the website’s interactive details and the way the product was presented, not only its appearance.
Sol also produced a website, an interactive user-interface mock-up, a marketing video and functioning software. Bowen liked the interactive interface but found the renders awkward and some elements offset. They thought the advertisement would have benefited from a voice. In the software test, Sol could not change volume by voice, although its settings panel offered controls.
Bowen called Sol a useful workhorse that helped port the robot software. When choosing between the complete product packages — software, site, branding and marketing — they picked Sonnet’s Perry result.
Free newsletter
Get the morning briefing
Start each day with the AI stories that matter and why — a short, free email from our newsroom.
The parking challenge had no clear winner
Bowen next had each model control a remote-control car using camera images and an Arduino-based setup, aiming to park between two other cars. Neither completed the parking.
Sol estimated movement from the camera images and developed a way to plan possible paths. Bowen was particularly impressed by how carefully it approached the task and how close the car came, even though it did not finish.
Sonnet tried combining steering and movement commands despite the instructions not confirming that they would work together. Bowen liked its willingness to try them rather than treat the uncertainty as a prohibition. Sonnet continued after Bowen said the car could be bump-parked.
Each model also made a motion graphic about its attempt. Bowen found Sonnet’s more visually elaborate, while Sol’s narrated graphic explained its method better. One moment drew a strong objection: Sonnet flagged Bowen’s request for a debrief as potential knowledge distillation, which Bowen called a major negative.
For the parking challenge as a whole, Bowen saw no clear winner or clear failure.
Sonnet’s music and video won decisively
For the last task, Bowen provided a roughly 50-second song, its separate audio tracks, lyrics and video clips. The request was to mix and master the music and make a promotional video.
Sol finished in under an hour and 40 minutes. Bowen liked a transition and some of its editing, but gave the mix their sharpest criticism of Sol: “That was a bad mix. I could have done better than that with Garage Band”. They also found its video less polished than Sonnet’s.
Sonnet’s result arrived after about two hours, by Bowen’s estimate. It added sounds, changed the instrumental and applied color treatment to the footage. Bowen found that it also handled a video overlay more consistently. In their view, Sonnet substantially improved both the music and the video, delivering a significantly better overall result.
Bowen speculated that Sol’s bare-minimum approach in some tasks might reflect a restriction on how much effort it applied. They still considered Sol a good model, but their preference for these tasks was decisive. Bowen said, “if I had to go do something right now, I would be using Sonnet”.