Google shipped Gemini 3.7 Flash on Thursday, three weeks after 3.6 Flash, at half the per-token price through year-end, and with no updated timeline for the delayed 3.5 Pro flagship that was supposed to anchor the family. The release is doing two jobs at once: pushing a workhorse coding model into the agent workflows where Anthropic has been eating Google’s lunch, and papering over a flagship gap that senior leadership can’t yet close.
The numbers are what Google wants you to look at. On DeepSWE v1.1, 3.7 Flash scores 65.3%, up from 3.6 Flash’s 49.0%. On FrontierCode 1.1 Main, a 100-task benchmark spanning multiple languages and grading style-guide compliance and bug testing, it lands at 43.6% versus 34.4% for the predecessor. Arena.ai’s WebDev Arena moves it from Elo 1538 to 1588. Google claims wins on nine benchmarks against comparable models from Anthropic and OpenAI, including 34% on a business-document benchmark called GDP.pdf where the reference points are Claude Sonnet 5 and GPT-5.6 Terra.
Introductory pricing is $0.75 per million input tokens and $3.75 per million output, roughly half of 3.6 Flash, expiring December 31, 2026. Context windows are unchanged at 1 million in and 64,000 out. The architecture, per the model card, inherits the transformer-based mixture-of-experts design from Gemini 3 Pro by way of 3.6 Flash.
“Gemini 3.7 Flash delivers a noticeably improved developer experience over 3.6 Flash. It thinks more diligently, putting in more effort into multi-step planning and tool calls,” said Tulsee Doshi, senior director of product management at Google. Read that as positioning: Reuters reports Google is pitching the model to businesses building autonomous agents that plan, call tools, and complete multi-step workflows. That’s the Claude-shaped hole in Google’s revenue story.
The pricing cut and the release cadence both make more sense once you look sideways at what isn’t shipping. Gemini 3.5 Pro, the delayed flagship, still isn’t in the Gemini app model picker, which on Thursday quietly swapped 3.6 Flash for 3.7 Flash alongside 3.5 Flash-Lite and 3.1 Pro. In July, Google told Reuters 3.5 Pro was being tested with partners and coming “soon.” That was a month ago. On the same July earnings call, Sundar Pichai defended the AI strategy against concerns Google had ceded ground on coding.
Bloomberg’s Micah Barkley notes the org chart is also in motion. Demis Hassabis stepped aside as DeepMind chief last week in favor of his deputy Koray Kavukcuoglu; two original technical co-leads of Gemini reportedly quit to co-found a startup; and Sergey Brin is said to be personally urging key AI staff to go all-in on Gemini. Founder-mode interventions and half-price Flash releases tend to arrive together for the same reason.
Spark, Google’s 24/7 personal agent, is now powered by 3.7 Flash across more than 160 countries via Google AI Pro and Ultra. It’s a real product update. It’s also what a company ships when the flagship isn’t ready and the coding narrative won’t wait.
Sources
- https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/
- https://www.bloomberg.com/news/articles/2026-08-13/google-debuts-new-gemini-flash-while-top-ai-model-still-delayed
- https://www.thestar.com.my/tech/tech-news/2026/08/14/google-unveils-gemini-37-flash-ai-model-for-coding-agent-workflows
- https://9to5google.com/2026/08/13/gemini-3-7-flash-launch/
- https://siliconangle.com/2026/08/13/google-launches-gemini-3-7-flash-coding-ai-agent-projects/