10:16AM: I've been thinking a lot about mixing R1 and other models recently. I've been doing this for quite a while now - using o1 pro to plan and o1 to execute. Sometimes I switch to Sonnet for execute phase, but my Asian brain wants to maximize utility of the $200 ChatGPT Pro subscription. The folks at Cline have added support for R1 + Sonnet directly into their tool. The results look interesting in their video. I certainly like how Sonnet is so much faster at execution on coding than o1 for what looks like ~ same quality. https://x.com/thankscline/status/1882217765043671323 Pietro Schirano has been doing great work at extracting reasoning tokens from an R1 response and using that to prompt Sonnet to generate the output. This is essentially the same idea as what the Cline folks just announced above. https://x.com/skirano/status/1882249162085016039