Gemini 3.7 Flash: Google Speeds Up Again and Halves the Price
Google announced Gemini 3.7 Flash on August 13, 2026, just three weeks after Gemini 3.6 Flash shipped. The Mountain View giant is keeping up a release cadence unmatched in the industry and is pushing hard on two fronts: coding performance, where the gains are dramatic, and price, halved compared with the previous model at its own launch. The model is already available to developers and is rolling out gradually in the Gemini app.
Three weeks between models: an unprecedented pace
Gemini 3.6 Flash was released on July 21, 2026. Twenty-three days later, Google follows up with version 3.7. The company presents this model as the direct result of developer feedback and algorithmic innovations it plans to reuse in future models.
A word on the range is in order. At Google, the Flash designation covers models optimised for cost-performance: faster and far cheaper than the Pro versions, they are built to absorb large request volumes. This is the workhorse of the line-up, the one companies deploy at scale.
Major coding gains
Code is where the progress is clearest. Google claims strong gains over Gemini 3.6 Flash on debugging and issue resolution. The figures put forward are as follows:
- DeepSWE v1.1: from 49.0 percent to 65.3 percent, more than 16 points gained.
- FrontierCode 1.1 Main: from 34.4 percent to 43.6 percent.
- Arena.ai WebDev Arena: an Elo score of 1,588, against 1,538 for the previous month’s model.
These three references measure different things. DeepSWE gauges a model’s ability to solve real software engineering problems. FrontierCode tests advanced programming tasks. WebDev Arena relies on blind comparisons carried out by users on web development tasks, with an Elo-style ranking borrowed from chess.
In web development, Google says the model generates more functional layouts and more feature-complete apps in fewer prompts. For interface generation, the company highlights high fidelity to the reference design supplied, whether that is a screenshot, an image or a full design system.
Complex documents and automation: the other gains
Beyond code, Google claims progress in knowledge-dense fields such as finance, law and biosciences.
On the GDP.pdf benchmark, which tests a model’s ability to process complex documents, Gemini 3.7 Flash reaches 34.0 percent against 22.0 percent for version 3.6. On AutomationBench, which measures the ability to complete real business workflows, the score moves from 17.0 percent to 30.4 percent. Those two increases, close to a doubling, matter more for enterprise use than purely conversational gains.
Google also stresses the model’s behaviour as an autonomous agent: it better adapts to roadblocks, clarifies intent when needed and follows instructions with greater fidelity. The company describes a model that puts more effort into multi-step planning and tool calls, which translates into less manual oversight and fewer retries.
Half the price
This is the most concrete argument for developers. Until the end of 2026, Gemini 3.7 Flash is offered at an introductory rate of 0.75 dollars per million input tokens and 3.75 dollars per million output tokens. That is half the price at which the previous model launched.
A token roughly corresponds to a fragment of a word: it is the standard billing unit for language models. In practice, a better model at half the cost changes the economics of many projects, particularly those processing large document volumes or running agents continuously.
One caveat: this is an introductory rate, valid until the end of the year. Google has not specified what pricing will apply afterwards.
Where to use it right now
In the Gemini app, the 3.7 Flash model is rolling out to Spark, Google’s personal agent, reserved for AI Pro and Ultra subscriptions. Google promises a more efficient agent for knowledge work, with improved tool use across Google Workspace apps.
For developers and businesses, the model is available in Google Antigravity, AI Studio, Android Studio, the Gemini Enterprise Agent Platform and the Gemini Enterprise app.
On safety, Google says it has updated safeguards against misuse in chemical, biological, radiological and nuclear domains, as well as cyber offence capabilities, while preserving legitimate use cases.
What it means for you
If you write software, the calculation is simple: a clearly better model for debugging, interface generation and agent execution, at half the price. Projects that stalled on cost per request deserve a fresh look.
If you simply use the Gemini app, the effect will be quieter and currently concerns AI Pro and Ultra subscribers via Spark. The gain will show up mainly on long tasks calling several tools.
One underlying question remains: at this pace, three weeks between models, the value of a benchmark ranking erodes quickly. These scores are reported by Google and had not been reproduced by independent third parties at the time of writing. As always, real-world testing will settle it.
Have you already tried Gemini 3.7 Flash on your own projects? Share your first impressions in the comments on Wanda-techs.com.
Read also:
Share this content:
Post Comment