Google has released a new "Flash" model roughly every three weeks this summer. Each one is a little cheaper, a little better at writing code, and arrives with a blog post insisting it's not a stopgap.
Meanwhile, the model everyone's actually been waiting for hasn't shown up at all.
What Actually Shipped
On August 13, 2026, Google released Gemini 3.7 Flash, positioned for coding, autonomous AI agents, and knowledge-intensive document work. It arrives just three weeks after Gemini 3.6 Flash.
FrontierCode 1.1 Main (production-quality code across languages, including bug testing and style-guide compliance): 43.6%, up from 34.4%
DeepSWE v1.1 (long-horizon software engineering): 65.3%, up from 49.0%
WebDev Arena Elo: 1588, up from 1538, the top score in Google's own comparison table
Google says the model outperformed comparable models from Anthropic and OpenAI across nine benchmarks it tested
Particularly improved at generating user interfaces that closely match reference images or design systems a developer uploads
Google's own model card describes 3.7 Flash as "a refinement of 3.6 Flash with algorithmic improvements to the core reasoning foundation, not a new pretraining run," and the company frames the pace itself as a feature: "a direct result of developer feedback and algorithmic innovations that we look forward to bringing to future models."
The Specs and the Price
Context window of up to 1 million input tokens, up to 64,000 output tokens
Accepts text, images, video, audio, code, and PDF files; generates text output
Knowledge cutoff remains March 2026, unchanged from 3.6 Flash
Introductory pricing: $0.75 per million input tokens, $3.75 per million output tokens, roughly half the prior model's rate, through December 31, 2026
Standard pricing doubles to $1.50 / $7.50 starting January 1, 2027
Available via Google AI Studio, the Gemini API, Google Antigravity, Android Studio, and, for enterprise, Gemini Enterprise and the Gemini Enterprise Agent Platform
Google is also rolling the model into Gemini Spark, its always-on personal AI agent for Pro and Ultra subscribers across more than 160 countries, which can work inside Gmail, Google Docs, and other Workspace tools under a user's direction. Google says the update makes Spark noticeably better at multi-step tasks across those apps.
| Benchmark | Gemini 3.6 Flash | Gemini 3.7 Flash |
|---|---|---|
| FrontierCode 1.1 Main | 34.4% | 43.6% |
| DeepSWE v1.1 | 49.0% | 65.3% |
| WebDev Arena (Elo) | 1538 | 1588 |
| Input price (per 1M tokens) | $1.50 | $0.75 (introductory) |
The Elephant Not in the Room
Here's what makes this release more interesting than a routine model update: it landed with no update at all on Gemini 3.5 Pro, the larger flagship model Google has been promising, and hasn't shipped.
Axios reported that Google wouldn't say what its plans for Gemini 3.5 Pro actually are, even as reporters asked directly around the 3.7 Flash launch. Reuters noted the same gap in its own coverage: Google "offered no details on when its flagship Pro model will be released."
The silence lands at a specific moment. CEO Sundar Pichai used Google's July earnings call to push back directly on concerns the company had fallen behind rivals in AI, particularly in coding, after delays to its flagship model. The 3.7 Flash release also comes on the heels of a leadership shake-up at Google DeepMind: Demis Hassabis handed day-to-day leadership to deputy Koray Kavukcuoglu, while the two engineers who had originally led Gemini's technical development left the company entirely to start their own venture, according to Reuters.
Gemini 3.7 Flash: FAQ
Gemini 3.7 Flash is a Google AI model released August 13, 2026, aimed at coding, autonomous AI agent workflows, and knowledge-intensive document tasks. It's the newest entry in Google's Flash tier, arriving just three weeks after Gemini 3.6 Flash, and is described by Google's own model card as a refinement of 3.6 Flash rather than a new pretraining run.
On Google's FrontierCode 1.1 Main benchmark, which tests production-quality code across multiple languages including bug testing and style-guide compliance, Gemini 3.7 Flash scored 43.6%, up from 34.4% for Gemini 3.6 Flash. On DeepSWE v1.1, a long-horizon software engineering evaluation, it scored 65.3%, up from 49.0%. Google also says the model outperformed comparable models from Anthropic and OpenAI across nine benchmarks it tested.
Introductory API pricing is $0.75 per million input tokens and $3.75 per million output tokens, half the rate of Gemini 3.6 Flash, available through December 31, 2026. Standard pricing will roughly double, to $1.50 and $7.50 respectively, starting January 1, 2027.
The model supports a context window of up to 1 million input tokens and can generate up to 64,000 output tokens. It accepts text, images, video, audio, code, and PDF files as input, and produces text output. Its knowledge cutoff remains March 2026, unchanged from Gemini 3.6 Flash.
As of the Gemini 3.7 Flash launch, Google had not released Gemini 3.5 Pro, a previously promised upgrade to its larger flagship model, and according to Axios, the company would not say what its plans for the model actually are. The release also follows a leadership shake-up at Google DeepMind, where Demis Hassabis handed day-to-day leadership to deputy Koray Kavukcuoglu, and two engineers who had led Gemini's technical development departed to start their own company.
Google says the release included automated testing, human-led red-teaming exercises, and its own Frontier Safety Framework assessments covering cyberattack and chemical, biological, radiological, and nuclear (CBRN) misuse risks. According to the model's safety card, Gemini 3.7 Flash did not reach a critical capability level in any tested risk area, though Google added extra safeguards in the CBRN and cyber domains regardless.
Jans Bock-Schroeder
Publisher & Founder of AI Angst
Coming from the world of art, photography, and the luxury market, Jans launched AI Angst in 2025 to explore the cultural, ethical, and psychological impacts of artificial intelligence. His work bridges creative vision with critical technology analysis, offering clarity in an era of rapid technological change.
Sources and Citations
This article is based on the following sources, published August 13, 2026:
-
Google: "Gemini 3.7 Flash: our most intelligent workhorse model" (August 13, 2026)
Primary official source for the model's stated capabilities, pricing, and Google's own framing.
https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/ -
9to5Google: "Gemini 3.7 Flash launches three weeks after last model, live in Spark" (August 13, 2026)
Source for the WebDev Arena Elo scores and Gemini Spark integration details.
https://9to5google.com/2026/08/13/gemini-3-7-flash-launch/ -
Quartz: "Google launches Gemini 3.7 Flash AI model for coding" (August 13, 2026)
Source for the Gemini 3.5 Pro delay, the Axios reporting, and the Google DeepMind leadership changes.
https://qz.com/google-gemini-37-flash-coding-ai-model-081326 -
SiliconANGLE: "Google launches Gemini 3.7 Flash for coding, AI agent projects" (August 13, 2026)
Source for the benchmark comparisons against Anthropic and OpenAI models, and UI-generation improvements.
https://siliconangle.com/2026/08/13/google-launches-gemini-3-7-flash-coding-ai-agent-projects/ -
Basic Tutorials: "Gemini 3.7 Flash Unveiled: New AI Model for Coding and Agents Costs Half as Much" (August 13, 2026)
Source for the safety-testing disclosures and comparison to OpenAI's delayed "Astra" model.
https://basic-tutorials.com/news/gemini-3-7-flash-unveiled-new-ai-model-for-coding-and-agents-costs-half-as-much/
Published: August 15, 2026. Sources verified at time of publication. All external links open in a new tab. Benchmark figures are Google's own reported results and have not been independently re-verified by a third party.
