@Alibaba_Qwen Qwen is the queen of intelligence per b parameters. They showed us this with their 3.6 27b, a small model that stomps Gemini pro often. It just shows that they are architected for efficiency and intelligence.
@johnnysn_sp@Alibaba_Qwen The main improvement isn't in knowledge density because it's probably the most knowledge dense model in the world. The improvements would come in thinking.
@thsottiaux Codex 5.3 is fucking AMAZING at coding. I really miss an agent that was seriously fast at straight code spitting execution in the GPT lineup. Back in 5.3 using 5.3 codex as the coder and 5.3 GPT as the dialectical agent was like pockets on a sundress.
@Jchammond_ I like using claude opus to plan mainly, also help with dev/ops stuff. I always run anything I plan through codex GPT 5.4 XHIGH (dialectical) and it's like calling in a parent that thinks you have failed at life compared to your surgeon brother when you present your shiny SDD.
@davepl1968 You can do almost all things you can do with openclaw with a script. It's kinda lame other than something fun to play with on airplanes lol.
@carlrichell Wait till John Q Uneducated public realizes that reviewing code takes almost as long as writing it, and fixing or directing the fixing of bad code takes exactly as long as writing it correctly.
Saying people can be replaced with an LLM(AI) is like saying a mechanic can be replaced by an impact driver that works 85 percent of the time. Anyone selling human replacements is a grifter.