Google released Gemini 3.7 Flash, a lightweight model for coding and agents. The official Google AI Studio account announced it at 2:07 a.m. KST on August 14.
What stands out more is the speed. It arrives just 3 weeks after its predecessor, 3.6 Flash.
Replacing 3.6 Flash After Just 3 Weeks
Google said this release is "a direct result of developer feedback and algorithmic improvements." Google DeepMind CEO Demis Hassabis also promoted it personally.
On his X account, Hassabis wrote, "a major upgrade for software engineering, web development, and knowledge work." He added that the introductory price is "half the original price of 3.6 Flash."
The Performance Gains in Numbers
| Benchmark | 3.6 Flash | 3.7 Flash |
|---|---|---|
| FrontierCode 1.1 Main | 34.4% | 43.6% |
| DeepSWE v1.1 | 49.0% | 65.3% |
According to MarkTechPost, the new model improved at debugging, issue resolution, and first-attempt code accuracy. It also reported stronger production-ready code generation.
In agent workflows, it plans and executes multi-step tasks more reliably. The company said the goal is to cut retries and human intervention when it hits obstacles.
The Price List: Now Is the Cheapest It Will Be
The introductory price is $0.75 per million input tokens and $3.75 per million output tokens.
But this price is temporary. From January 1, 2027, it rises to $1.50 for input and $7.50 for output. (Source)
In effect, you can use it at half the price of 3.6 Flash through the end of the year.
Where You Can Use It
It is available right away in the Gemini API, Google AI Studio, Android Studio, Google Antigravity, and enterprise platforms. Logan Kilpatrick of Google AI Studio shared links to the playground and build environments himself.
Gemini 3.7 Performance Figures Are Self-Reported
The per-token prices and benchmark numbers were cross-checked against press reports. However, the following are parts that still need to be checked against the official documentation.
- Support for on-device or offline execution was not confirmed in the official announcement.
- Benchmark results by language, including Korean, have not been published separately.
If your team plugs a model into a coding agent, now is a good time to run a migration test before the year-end price increase.