GPT-6.1 Sol in practice: strong redesign, uneven game results

Robbert van Empel’s September 30 video takes GPT-6.1 Sol beyond OpenAI’s launch claims and tests it on two creative coding tasks. The video compares the model with GPT-6 Sol and GPT-6 Astra, then asks what its lower listed price means when a user asks it to build a website and a small game. This is an individual demonstration, not a controlled benchmark.

For context, OpenAI lists GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens, against $10 and $50 for Astra. Van Empel also shows Artificial Analysis figures available during the video: an intelligence score of about 52 for Sol and 53 for Astra, with estimated task costs of $0.72 and $3.26 respectively. Those scores are index results, not percentages of human or model intelligence, and cost per task varies with the work and settings. The video’s figures are a snapshot, while OpenAI’s own evaluations remain vendor-reported.

Van Empel’s first hands-on task is an interactive redesign of a Wikipedia page. He finds the Sol 6.1 result a clear improvement over Sol 6 and praises its appearance and animation. Some interactions do not work as intended: the paper roll cannot be freely used and the animation’s cat test has no cat. He says the result resembles Astra’s version and places it near the top of his comparison, while preferring Claude Sonnet 5.5’s redesign.

The second prompt creates a South Park-themed LEGO globe. In the video, Sol 6.1 produces a recognizable world, character dialogue, humor and working movement physics. It can pick up a brick, but stacking bricks does not work; Van Empel considers Astra stronger on this task. Because the prompt uses recognizable copyrighted characters, the demonstration is not a legal assessment of what users may publish.

The video also relays other creators’ reported costs for short animations, which they say were lower with Sol than with competing models. Those examples use different tasks and are not a standardized comparison. The practical conclusion is narrower than “Astra for one-fifth the price”: Sol 6.1 can produce polished creative prototypes at lower token prices, but results depend on the task and still need review. The video gives buyers concrete examples to test against their own workloads.