Weekly Summary #ArtificialIntelligence and
Holy sh*t this might kill the $200/mo AI subscription
Someone replaced Opus 5 with DeepSeek V4 Flash and figured out how to run multiple models as subagents inside Codex the insane part is the setup takes 3 minutes 1. install model-router 2. enable Subagent Models → All
Deepseek V4 flash just beat its own pro model on all 9 agent benchmarks.
And the biggest upgrade had nothing to do with making the model bigger. What changed: → DeepSWE jumped from 7.3 → 54.4 after retraining → Terminal Bench 2.1: Flash scored 82.7 vs Pro at 72.1 → Same
Use DeepSeek-V4-Pro-0813 to ask your gf out for hotpot
Just hit Zinley. cleared a work pile with it, then had it ping her about dinner. both stuck. i barely watched. web, desktop, tele, iMessage. dump the job and leave.
You have reached the end of the archive
All of deepseek