Gemini 3.7 Flash

What's this about?

Google released Gemini 3.7 Flash, which it calls its most intelligent workhorse model yet for coding and agents — only three weeks after Gemini 3.6 Flash. It scores 56 on the Artificial Analysis intelligence index, up from 52, and beats Claude Sonnet 5 on nearly every benchmark at less than half the price. The pricing is introductory: $0.75 per million input tokens and $3.75 per million output through the end of 2026, after which it doubles.

What exactly is a "workhorse" or "Flash" model?

Labs ship model tiers the way car makers ship engine sizes. The top tier is the most capable and most expensive, used for the hardest reasoning. A "Flash" or workhorse tier is deliberately smaller and faster, meant for the enormous volume of ordinary requests — summarizing, classifying, routine code edits — where paying frontier prices makes no economic sense. The interesting shift is that these cheap tiers now score close to what the top tier scored a year ago, so for most production work the workhorse is what actually gets deployed.

What's the community saying?

1. The comparison Google chose was immediately challenged. Google benchmarked against GPT-5.6 Terra, but commenters pointed out Terra is roughly half the price, and that the recent Grok 4.6 appears to be both better and cheaper — leading to the thread's sharpest question, "hard to understand why anyone would choose 3.7 Flash. Is DeepMind still a frontier lab?"
2. It burns more tokens to get its score. A careful commenter noted the index rose from 52 to 56, but output tokens per task went from 26k to 37k. Because the price was halved, cost per task still falls — but the intelligence gain is partly bought with verbosity.
3. The introductory pricing is a puzzle. Doubling on December 31 implies Google expects people still using it in five months, which several found odd given 3.6 Flash was only three weeks old. Simon Willison made the same point publicly.
4. A real defense of the model. A well-received counterpoint: beating Sonnet 5 on almost all benchmarks at under half the price is genuinely competitive, and it puts Google back in the game — just no longer leading the frontier.
5. "Where's the Pro model?" Repeated asks about when another Gemini Pro arrives, with suspicion that Google is retreating into the fast-but-moderate niche. One sardonic take: "another failed 3.5 Pro run rebranded as 3.7 Flash."
6. The text-only use case is contested. For pure text, commenters argued DeepSeek V4 Flash/Pro is 13–26x cheaper at comparable intelligence and available from many providers — leaving multimodal as 3.7 Flash's clearest advantage.