Google launches Gemini 3.7 Flash just three weeks after its predecessor
The new model brings measurable gains in coding and document reading, and arrives with a cut-price introductory rate as Google feels pressure from cheaper rivals.
Key points
- Google released Gemini 3.7 Flash today, replacing Gemini 3.6 Flash, which launched only three weeks earlier.
- On the FrontierCode 1.1 coding test, Gemini 3.7 Flash scored 43.6%, up from 34.4% for 3.6 Flash.
- On the DeepSWE v1.1 software-engineering benchmark, the new model scored 65.3%, compared with 49% for its predecessor.
- Business-workflow performance, measured by the AutomationBench test, jumped from 17% to 30.4%.
- Google is offering an introductory lower price to compete with cheaper models from rival labs.
Google has released a new version of its Gemini Flash model, the AI (artificial intelligence) system it positions as its everyday workhorse for developers and businesses. Gemini 3.7 Flash is now rolling out across Google's platforms, replacing Gemini 3.6 Flash, which itself only arrived three weeks ago. We first covered the 3.7 Flash line on 13 August 2026.
The quick turnaround is deliberate. Google says the new model is the result of core engineering improvements and developer feedback. The company's also cutting the introductory price, a move aimed squarely at competitors offering cheaper models. Our 6 August piece on AI pricing instability found that budgets are already blowing up before businesses notice the bill, which makes that introductory rate more than a marketing gesture.
What actually got better?
Coding is the headline improvement. On FrontierCode 1.1 Main, a standard industry test that measures how well a model can write and fix software code, the score rose to 43.6% from 34.4%. DeepSWE v1.1, a separate benchmark for real-world software engineering tasks, climbed from 49% to 65.3%. Those are meaningful gaps, not rounding errors.
Document reading also improved. The GDP.pdf benchmark, which measures how well a model processes complex documents, rose from 22% to 34%. That's a practical gain for anyone using AI to summarise contracts or research papers. Senior Director Tulsee Doshi also points to the WebDev Arena score, a community leaderboard where real users rate AI-generated web designs, rising from 1,538 to 1,588.
What does this mean for people using Google's AI tools?
If you use Gemini for coding or long documents, you should see real improvements. The upgrade happens in the background; you don't need to do anything.
For businesses running automated tasks, the AutomationBench result is the one to watch. That test measures how reliably a model handles common office workflows, and the score nearly doubled, from 17% to 30.4%.
| Benchmark | Gemini 3.6 Flash | Gemini 3.7 Flash |
|---|---|---|
| FrontierCode 1.1 Main | 34.4% | 43.6% |
| DeepSWE v1.1 | 49.0% | 65.3% |
| GDP.pdf (document reading) | 22.0% | 34.0% |
| AutomationBench | 17.0% | 30.4% |
| WebDev Arena score | 1,538 | 1,588 |
Notably absent from today's announcement, as Ars Technica flagged, is Gemini 3.5 Pro, the more powerful model developers have been waiting for. Flash models are built for speed and cost, not maximum capability. The bigger release is still to come.
What happens next?
Gemini 3.7 Flash is available now. No date has been given for the higher-end Gemini Pro update, but the pace of Flash releases tells you Google's feeling real competitive heat.
