Google has released Gemini 3.7 Flash, a new AI model for software development, agent tasks, and document processing. The update was released on August 13, 2026—just three weeks after Gemini 3.6 Flash—and shows significant improvements across several benchmarks. Nevertheless, the introductory price is half that of its predecessor.
Key Points at a Glance
- Release: August 13, 2026, just three weeks after Gemini 3.6 Flash
- Significant improvements in coding (FrontierCode), agent tasks (DeepSWE), web development (WebDev Arena), and document comprehension (GDP.pdf)
- Introductory price through December 31, 2026: $0.75 per 1 million input tokens, $3.75 per 1 million output tokens; thereafter, $1.50 and $7.50, respectively
- Context window of up to 1 million tokens, maximum output of 64,000 tokens
- Available via Google AI Studio, Android Studio, Google Antigravity, the Gemini Enterprise Agent Platform, and Gemini Spark for AI Pro and Ultra subscribers
- As of March 2026; in some areas, only through January 2025
What’s new in Gemini 3.7 Flash?
In its announcement, Google calls Gemini 3.7 Flash the most powerful Flash model to date for coding and agents. According to the company, it notably improves the model’s ability to handle unexpected obstacles and execute instructions more reliably. For multi-step tasks, the model is designed to execute tools more independently, requiring less manual oversight from humans. Google also reports progress in understanding complex documents and automating business processes in fields such as finance and law.
According to 9to5google, this rapid pace is “a direct result of developer feedback and algorithmic innovations.” Google is also promoting Gemini 3.7 Flash for Google Workspace: The personal assistant Gemini Spark is expected to become more efficient at office tasks as a result.
Benchmarks Compared to Gemini 3.6 Flash
These figures come from Google’s own tests; independent third-party retests are not yet available.
| Benchmark | Gemini 3.6 Flash | Gemini 3.7 Flash |
|---|---|---|
| FrontierCode 1.1 Main (Code Quality) | 34.4% | 43.6% |
| DeepSWE v1.1 (Software Engineering) | 49.0% | 65.3% |
| WebDev Arena (Elo rating) | 1538 | 1588 |
| GDP.pdf (Document Comprehension) | 22.0% | 34.0% |
| AutomationBench (Business Workflows) | 17.0% | 30.4% |
The jump in performance for DeepSWE and AutomationBench is particularly striking: Both benchmarks test how well a model operates autonomously across multiple steps, rather than simply completing individual lines of code. This is precisely what Google is aiming for with the term “agent model.”
Pricing and Availability
An introductory price is in effect through December 31, 2026, which Google says is 50 percent below the launch price of Gemini 3.6 Flash. After that, the regular pricing tier will apply:
| Period | Input (per 1 million tokens) | Output (per 1 million tokens) |
|---|---|---|
| Introductory Price (through Dec. 31, 2026) | $0.75 | $3.75 |
| Regular price (effective Jan. 1, 2027) | $1.50 | $7.50 |
There are several ways to access the model: Developers can access Gemini 3.7 Flash via Google AI Studio, Android Studio, and Google Antigravity; businesses can access it via the Gemini Enterprise Agent Platform and the Gemini Enterprise App. Individual users with a subscription to Gemini AI Pro or AI Ultra can access the model through their personal assistant, Gemini Spark.
Technical Specifications
- Context window: up to 1,000,000 tokens
- Maximum output length: 64,000 tokens
- Input: Text, images, audio, and video
- Output: Text
- State of knowledge: March 2026; for certain areas, only up to January 2025
Regarding security, Google cites automated testing, human-led red-team exercises, and its own Frontier Safety assessments in areas such as cyberattacks and chemical, biological, radiological, and nuclear (CBRN) risks. According to the model map, Gemini 3.7 Flash behaves similarly to its predecessor in terms of tone and safety, with few unnecessary refusals to answer. The model did not reach a critical capability level in any of the tested areas, which is why Google has implemented additional safeguards against misuse in the CBRN and cyber domains.
Three weeks after the last update: Why so soon?
Gemini 3.6 Flash was released as recently as July 21, 2026. The fact that Google is following up with 3.7 Flash just three weeks later fits a pattern: Google is noticeably picking up the pace with the smaller, more affordable Flash models, while the larger flagship model, Gemini 3.5 Pro, is reportedly still a long time coming, according to Bloomberg and Axios. Google itself has not commented on the reasons for the delay.
The contrast with the competition is striking: Just in early August, OpenAI postponed the release of its next major model, named Astra, because internal security tests could not rule out a critical rating regarding its cyber capabilities. Google is taking the opposite approach with the smaller Flash model: rapid, iterative updates instead of a single major leap.
Where Gemini Stands Today
Even though Gemini 3.7 Flash is initially aimed at developers and businesses, Google’s own hardware demonstrates just how deeply the model family is now integrated. The recently unveiled Pixel Watch 5 processes some Gemini queries directly on the watch, and on the Pixel 11 Pro XL, a dedicated LED bar called HiLight even indicates when Gemini is active. Improvements from the Flash series are likely to make their way into these consumer products over time, even though 3.7 Flash primarily serves developer tools and enterprise customers at launch.
Conclusion
Gemini 3.7 Flash is not a new flagship release, but rather a targeted update for coding, agent tasks, and document processing. According to Google’s own figures, it delivers performance gains in all five tested benchmarks while simultaneously cutting the launch price in half. Whether the improvements in FrontierCode, DeepSWE, and similar tools will be just as noticeable in real-world developer workflows cannot yet be conclusively assessed based on the figures currently available, which were reported by Google itself. For teams already working with Gemini models, the switch is likely a no-brainer due to the lower price. Those waiting for the next major model will have to be patient: Gemini 3.5 Pro is still a long way off.
