Raw GPT-5.6-Sol Traces Show Concise Reasoning
Posts share screenshots of raw GPT-5.6-Sol traces solving geometry problems with vectors and Fermat points.
Pseudonymous account scaling01 posted screenshots of decoded internal reasoning from GPT-5.6-Sol on geometry tasks involving Fermat points, vectors, and the law of cosines. The traces follow a concise style likened to an Office meme about using few words. Replies note this style appeared with GPT-5.5 and that the model tests multiple hypotheses until reaching a solution. One comparison finds GPT-5.6-Sol more efficient than Claude Opus 4.8 traces despite operating at a lower abstraction level. No OpenAI confirmation appears in the posts.
raw GPT-5.6-Sol thinking traces OpenAI literally did The Office meme: "Why use many words when few words do trick"
Combined views
Raw GPT-5.6-Sol Traces Show Concise Reasoning
Posts share screenshots of raw GPT-5.6-Sol traces solving geometry problems with vectors and Fermat points.
Pseudonymous account scaling01 posted screenshots of decoded internal reasoning from GPT-5.6-Sol on geometry tasks involving Fermat points, vectors, and the law of cosines. The traces follow a concise style likened to an Office meme about using few words. Replies note this style appeared with GPT-5.5 and that the model tests multiple hypotheses until reaching a solution. One comparison finds GPT-5.6-Sol more efficient than Claude Opus 4.8 traces despite operating at a lower abstraction level. No OpenAI confirmation appears in the posts.
Reactions from ranked influencers
15 poststhis is beautiful
@scaling01 This evolved without any involvement from OpenAI as far as I know, and has been showing up in the rare raw CoT leaks from chat for quite a while now, since earlier in the year.
@AndrewCurran_ this very concise reasoning style only emerged with GPT-5.5 probably because it's a larger model and doesn't need all the filler words but yeah, we have seen some reasoning traces before and OpenAI models have always been most neuralese like
I looked at all the reasoning traces from Opus 4.8 and GPT-5.6-Sol to me it seems like Opus 4.8 is working on a much higher abstraction level, while Sol is just doing the dirty work. it just explores and keeps spamming new hypotheses until it finds a solution not only is Sol…
raw GPT-5.6-Sol thinking traces OpenAI literally did The Office meme: "Why use many words when few words do trick"
imo safer, more faithful: Anthropic more scalable: OpenAI
