Google DeepMind · 2026-08-13 · major
Gemini 3.7 Flash — Google's coding and agent workhorse at half the price
Gemini 3.7 Flash scores 65.3% on DeepSWE v1.1, up from 49.0% for Gemini 3.6 Flash, and arrives just three weeks later. Google charges $0.75 per 1M input tokens through 2026, half the older model's launch price.
Google's new Flash model gains 16 points on long-horizon coding work and launches at half the price of the model it replaces.
Key specs
| Deep swe v1.1 | 65.3% |
|---|---|
| Web dev arena elo | 1,588 |
Quick facts
| Maker | Google DeepMind |
|---|---|
| Price (input) | $0.75 / 1M tokens |
| Price (output) | $3.75 / 1M tokens |
| Standard price | $1.50 in / $7.50 out from Jan 1, 2027 |
| Modalities | Text, code, images, PDF documents |
| Availability | Gemini API, AI Studio, Android Studio, Antigravity, Gemini Enterprise |
| Consumer access | Gemini Spark for AI Pro and Ultra in 160+ countries |
Benchmarks
Pricing
| Input · Introductory rate through Dec 31, 2026 | $0.75 / 1M tokens |
|---|---|
| Output · Introductory rate through Dec 31, 2026 | $3.75 / 1M tokens |
| Input (standard) · From Jan 1, 2027 | $1.50 / 1M tokens |
| Output (standard) · From Jan 1, 2027 | $7.50 / 1M tokens |
What is it?
Gemini 3.7 Flash lifts Google's Flash-tier coding scores by a wide margin, reaching 65.3% on DeepSWE v1.1 against 49.0% for Gemini 3.6 Flash. The model handles text, code, images, and PDF documents. It landed on August 13, 2026, three weeks after 3.6 Flash, and is generally available from day one.
How does it work?
Algorithmic improvements from teams across Google DeepMind drive the gains, not a larger model. Google says the model thinks more diligently and puts more effort into multi-step planning and tool calls, which is what moves the agent numbers. The jump is largest on long-horizon work: enterprise workflow automation nearly doubles on AutomationBench, and expert PDF comprehension on GDP.pdf climbs from 22.0% to 34.0%.
Why does it matter?
Cheaper tokens plus higher coding scores change what a production agent can afford to run on. At $0.75 per 1M input tokens, Gemini 3.7 Flash costs half of what 3.6 Flash did at launch while scoring higher on every benchmark Google published. Teams that pushed hard coding work up to a Pro-tier model can now test whether the Flash tier covers it.
Who is it for?
Developers and enterprises running coding and agent workloads
Frequently asked questions
- How much does Gemini 3.7 Flash cost?
- Gemini 3.7 Flash costs $0.75 per 1M input tokens and $3.75 per 1M output tokens as an introductory rate through December 31, 2026. Google describes that as half the price Gemini 3.6 Flash launched at. From January 1, 2027 the standard rate rises to $1.50 input and $7.50 output per 1M tokens.
- How does Gemini 3.7 Flash compare to Gemini 3.6 Flash?
- Gemini 3.7 Flash beats Gemini 3.6 Flash on every benchmark Google published: 65.3% versus 49.0% on DeepSWE v1.1, 43.6% versus 34.4% on FrontierCode 1.1 Main, 34.0% versus 22.0% on GDP.pdf, and 30.4% versus 17.0% on AutomationBench. On WebDev Arena it scores 1588 Elo against 1538.
- Where can I use Gemini 3.7 Flash today?
- Gemini 3.7 Flash is generally available to developers through the Gemini API in Google AI Studio, Android Studio, and Google Antigravity. Enterprises get it in the Gemini Enterprise Agent Platform and the Gemini Enterprise app. Consumers reach it through Gemini Spark, the personal agent for Google AI Pro and Ultra subscribers in 160+ countries.
- Can I download the weights for Gemini 3.7 Flash?
- No. Gemini 3.7 Flash is a closed-weight Google model served only through the Gemini API and Google's own products. There is no open-weights release and no self-hosted option, so every request runs on Google infrastructure and is billed per token at the published API rates.
- What safety changes shipped with Gemini 3.7 Flash?
- Google updated the safeguards in Gemini 3.7 Flash against misuse in chemical, biological, radiological, and nuclear (CBRN) domains and in cyber offense, while keeping legitimate uses in those fields working. The safety work shipped alongside the coding and agent improvements rather than as a separate release.
Try it
https://aistudio.google.com/