On August 13, Google released Gemini 3.7 Flash, three weeks after 3.6 Flash. The release was quiet for a Google product. No grand stage. No Sundar Pichai keynote. Just a blog post and an API update.
Gemini 3.7 Flash is real. The improvements are real. The pricing is aggressive. But the context around it is impossible to ignore.
FrontierCode Jumped 9 Points. DeepSWE Jumped 16.
Google claims Gemini 3.7 Flash is its most capable "workhorse" model yet, optimized for programming, agent workflows, and knowledge work.
The benchmarks show meaningful jumps:
Benchmark | Gemini 3.6 Flash | Gemini 3.7 Flash | Gain |
|---|---|---|---|
FrontierCode 1.1 Main | 34.4% | 43.6% | +9.2 pp |
DeepSWE v1.1 | 49.0% | 65.3% | +16.3 pp |
GDP.pdf | 22.0% | 34.0% | +12 pp |
AutomationBench | 17.0% | 30.4% | +13.4 pp |
WebDev Arena Elo | 1538 | 1588 | +50 |
At $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026, the pricing is half the original list price of Gemini 3.6 Flash. A price that low is designed to do one thing: get developers to try it.
Independent testing by Artificial Analysis gives Gemini 3.7 Flash a composite Intelligence Index score of 53, ranking it around 17th among all models. Output speed hits roughly 340 tokens per second, top among comparable models.
36% Faster, 67% Accurate—Enterprises Are Noticing
Box, an enterprise content platform, ran Gemini 3.7 Flash through its complex work evaluation—realistic business tasks across finance, legal, healthcare, and consumer goods. The results: Gemini 3.7 Flash scored 67% accuracy versus 62% for Gemini 3.6 Flash, and completed tasks 36% faster (73 seconds vs 115 seconds). Financial services tasks saw the largest gains, jumping from 65% to 83%.
The combination—more accurate and faster—matters for enterprises. A model that is both cheaper and more capable at the same time is rare. For teams building agent systems, the marginal cost of each failure drops. Fewer retries. Less debugging. More work actually getting done.
The Pro Model Still Isn't Here
But here is the part Google is not emphasizing.
Gemini 3.5 Pro was promised in June. It is now August. The model is still not here. Meanwhile, Flash has cycled from 3.5 to 3.6 to 3.7 in the same period.
Reuters reported on August 13 that internal testing showed Gemini's new flagship version still lagged in coding performance, and Google delayed the release by two months. The same day, Google released 3.7 Flash. The timing is not a coincidence.
The flagship is stuck. Flash is moving. But Flash is not a replacement for the flagship. It is a workhorse, not a frontrunner.
This Is Kavukcuoglu's First Product. Brin Wanted It Faster.
The subtext of the 3.7 Flash release is Google DeepMind's leadership overhaul on August 5. Demis Hassabis stepped down as CEO; Koray Kavukcuoglu took over. Jeff Dean left to start a new company.
According to Reuters, Sergey Brin has been directly intervening in Gemini development for months. In April, as Anthropic previewed Claude Mythos, Brin addressed hundreds of DeepMind employees and urged them to move faster. He has been pushing resources toward specific directions and backing Kavukcuoglu's rise.
The 3.7 Flash release is Kavukcuoglu's first product under his new title. It is fast and cheap, exactly what a product-led turnaround looks like. But it is not what the market is waiting for. The market is waiting for the flagship.
Gemini 3.7 Flash is a solid product. For developers building agents on a budget, it might be the right choice. For Google, it buys time. It keeps developers in the Gemini ecosystem while the Pro team works through whatever is holding it back.
But the gap is growing. Flash is fast, cheap, and smart enough. The flagship is delayed, struggling, and not here yet. Google is competing with its own B-team while its A-team tries to catch up to Anthropic and OpenAI.
P.S. The introductory pricing ends January 1, 2027. By then, Google's pricing will double. If the Pro model still isn't out by then, the question changes: do developers keep building on Flash, or do they conclude Google is not serious about the front of the pack?
Frequently Asked Questions
Q: What is Gemini 3.7 Flash?
A: Gemini 3.7 Flash is the latest iteration in Google's "Flash" model line, released on August 13, 2026. It is optimized for programming, agent workflows, and automation, with improved coding and reasoning capabilities at a lower price.
Q: How much does Gemini 3.7 Flash cost?
A: Through the end of 2026, the price is $0.75 per million input tokens and $3.75 per million output tokens—half the original list price of Gemini 3.6 Flash. The price will double to $1.50/$7.50 on January 1, 2027.
Q: How much did Gemini 3.7 Flash improve over Gemini 3.6 Flash?
A: FrontierCode 1.1 Main jumped from 34.4% to 43.6% (+9.2 points). DeepSWE v1.1 jumped from 49.0% to 65.3% (+16.3 points). GDP.pdf went from 22% to 34%. AutomationBench from 17% to 30.4%.
Q: Where does Gemini 3.7 Flash rank among other AI models?
A: According to Artificial Analysis, Gemini 3.7 Flash has a composite Intelligence Index score of 53, ranking it approximately 17th among all models. It leads in output speed at roughly 340 tokens per second.
Q: What is the context behind this release?
A: Gemini 3.5 Pro was promised in June but still hasn't arrived. The 3.7 Flash release comes three weeks after 3.6 Flash and follows a major Google DeepMind leadership overhaul on August 5, with Sergey Brin directly pushing the team to accelerate.
Q: Who is Koray Kavukcuoglu?
A: Kavukcuoglu took over as CEO of Google DeepMind on August 5, replacing Demis Hassabis. He had been DeepMind's CTO and the primary liaison with Google Cloud. Gemini 3.7 Flash is his first product under the new title.
