GPT-5.6 Sol

GPT-5.6 Sol is one of the most closely examined models on ToTheMoon. In the summer of 2026 it turns up in almost every episode, because Alexander Volchek was running it on real projects rather than demos. Talking about it means talking about the price of access to the strongest models: how many tokens autonomous work burns and why a strong model can still disappoint as a product.

The first impressions, in July 2026, were mixed. In episode 134 the host found that its questions were more substantial than those of GPT-5.5 High, but that it over-complicated the task and had to be stopped by hand. In episode 135 GPT-5.6 Sol, alongside Claude Fable, was moving a large working site into a custom content-management system; the model could run for days and spawn parallel streams, yet duration alone proved nothing about quality. Episode 136 added slowness, extra steps and unreliable completion.

The assessment then hardened. In episode 138 Alexander described Codex with GPT-5.6 Sol burning tens of millions of tokens, spending seven days trying to build components and idling half the time, while a switch to 5.5 High produced results at once. Episode 139 put it plainly: “Once again, GPT-5.6 Sol did not produce a meaningful result for me” (episode 139), with the caveat that the model would improve over time. Episode 141 gave a figure: of forty billion tokens, roughly thirty-five had gone to Sol. Still, episode 143 credited the Pro version with going deep into search and analysis, and by episode 154 the conclusion was that a new version is not always better than the old.