Theo Browne says earlier Flash versions were strong but calls Flash 3 a decline and Flash 3.5 unsatisfactory in an X thread on AI models and lab principles

Flash 3.5 costs 3x more per token than earlier versions and burns far more tokens overall, making it 2x pricier to run than Gemini 3.1 Pro on the same tasks. It also scores worse than GPT-5.5 Medium on benchmarks while delivering no meaningful real-world gains, so Theo sees no viable use case. ^¹

He calls it a “useless model that should not be used for anything.”