For what it’s worth, I decided to give Google Gemini 3.8 a try this morning. And would you believe it, Gemini basically busted through major blockers that both Fable 5.1 Ultra and SOL Ultra had been bogged down on for a month — Fable 5 before it.
So it’s worth being precise about what was stuck and what got it moving.
What got it moving was a model Google positions as a workhorse. Gemini 3.8 Flash came out on Wednesday, the fourth Flash model Google has shipped in under four months by Artificial Analysis’s count, and on high reasoning it “averages ~300 output tokens per second.” My own read on SOL and Fable is around 60, so Gemini is running several times faster than the models it just outworked.
Speed alone wouldn’t mean much if the work were bad. Upon reviewing the work — which only took an hour for Gemini — both Claude and Codex declared it a major leap forward. They also discovered a bunch of errors. Gemini fixed them with dispatch.
That sequence matters more than the headline. The fast model built, two slower models from other companies judged, and the fast model corrected.
Why the cheap, fast model cleared what the expensive, deliberate ones couldn’t is an open question. It may be that a month of one family’s thinking had worn a groove, and a fresh family simply didn’t fall into it. That’s a guess.
What isn’t a guess is how it feels. Gemini works at 300 tokens per second, compared with SOL and Fable at around 60. It feels like a speed demon.
