Briefs · Emerging Technologies
Google Shipped a New Flagship Three Weeks After the Last One
Google shipped a new flagship model three weeks after the last one, and the difference is not the model. It is the thinking.
Gemini 3.7 Flash, released 13 August, is Google’s “most intelligent workhorse model yet for coding and agents”. It is not a new pretraining run, but rather algorithmic improvements to the Gemini 3.6 Flash reasoning foundation — the same base, the same training data, three weeks of post-training apart.
The gains are concentrated where developers live. DeepSWE, a long-horizon coding benchmark, jumps from 48.6% to 65.3%; FrontierCode rises from 34.4% to 43.6%, ahead of Claude Sonnet 5 and GPT-5.6 Terra. The trick is hybrid reasoning: a configurable thinking budget that spends reasoning tokens only where the problem demands them, across a million-token context window.
Conclusion: 3.7 Flash is not a bigger brain, but rather a cheaper thinker — and in the agent economy, cheaper thinking is the whole game.
Read the original at NYU Shanghai RITS
Summary in our own words. The original belongs to its publisher.