After gaining access to GPT-6.2 Sol, the tester tried it on roughly eight to ten projects and came away impressed, particularly by its visual and creative capabilities. It performed well at 3D work and could produce videos using code that looked like polished AI-generated footage rather than visibly scripted output. A standout result was a black-hole scene reminiscent of Interstellar, described as one of the tester’s best creations. Overall, Sol was judged competitive with Astra and Opus 5.5, though Opus remained stronger.
Speed, previously a major weakness of Sol, had improved substantially, and the model seemed more efficient. However, it still fell short of Opus 5.5’s performance and could be faster, with Google’s models offered as a benchmark. The tester also felt Sol retained a recognizable “GPT” aesthetic that could use refinement. Coding remains an open question: the code examined so far was not especially easy for humans to read, so more testing is planned in that area. Sol was still in limited “grey” testing, but the early feedback appeared broadly positive.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.