Introducing Gemini 3.7 Flash

Source: DeepMind Date: 2026-08-13 URL: https://deepmind.google/blog/introducing-gemini-3-7-flash/

Summary

Google shipped Gemini 3.7 Flash on 2026-08-13, three weeks after 3.6 Flash, positioned as its “most intelligent workhorse model yet for coding and agents.” Vendor-reported benchmark jumps over 3.6 Flash: FrontierCode 34.4→43.6, DeepSWE 49.0→65.3, GDP.pdf (document comprehension) 22.0→34.0, AutomationBench 17.0→30.4, WebDev Arena Elo 1538→1588. Introductory pricing is $0.75/$3.75 per million input/output tokens (half of 3.6 Flash) through end of 2026, available in AI Studio, Android Studio, Antigravity, Gemini Enterprise, and Gemini Spark.

Implications

Two-clocks closed-vs-open capability thread: a three-week release cadence for a coding/agent-focused mid-tier model is itself the signal — Google is iterating its cheap workhorse tier roughly as fast as the open-weight labs ship new checkpoints, which narrows the “closed labs move slow, open moves fast” framing that’s held for most of this year. The DeepSWE and FrontierCode gains land in the same benchmark family as DeepSeek-V4-Pro’s coding/agent jumps reported this week, making a direct (if vendor-reported, unverified) comparison point across the closed/open line. Pricing halved at the same time capability rose is also notable for the AI-financing thread — compute cost per unit of coding capability keeps falling on the vendor side, the opposite of the funding-pressure narrative elsewhere in the ecosystem this week.