Gemini 3.7 Flash was released on August 13 and was made generally available in more than 160 countries on day one. The model supports up to a million input tokens and returns 64,000 output tokens, reflecting its headline input and output capacities. It can read images, video, audio and PDFs, and it can call tools and drive a computer as part of its supported media and interface capabilities.
These points record the release date and day-one global availability alongside the model’s headline technical capabilities. The section provides a concise factual overview of Gemini 3.7 Flash’s initial availability and primary interface features.
Gemini 3.7 Flash can read images, video, audio and PDFs, enabling the model to accept and process visual, auditory and document inputs across those media types. The model supports up to a million input tokens and returns 64,000 output tokens, presented as its input and output capacity limits. It can call tools and drive a computer, functioning with tooling and interface capabilities that allow it to interact with external tools and systems. These capabilities are described together as the model’s headline functional features.
Google’s benchmark sheet places Gemini 3.7 Flash ahead of Claude Sonnet 5 and GPT-5.6 Terra on 11 of 18 tested categories. The benchmark comparison covers 18 tested categories in total. The article lists two headline metrics for Gemini 3.7 Flash: 1,588 Elo on Code Arena’s web development board and 30.4% on AutomationBench. Those figures are presented in the article as the headline numbers for the model.
This summary reports the benchmark positioning and the headline metrics noted in the article. It does not include testing procedures or qualitative assessments here.
Google measured zero-shot code generation using a single prompt to produce a browser game with no follow-ups. That run took 2 minutes and 13 seconds to pass. The test result recorded that the model produced a playable browser game on that run.
The game was playable on the first run; the syntax was clean, the collision and scoring logic held, and the visual quality sat above the price tier. Gemini 3.6 Flash could not produce a working file because its HTML was malformed. DeepSeek found 11 bugs in Gemini 3.6 Flash and shipped 8 fixes to make that earlier release playable.
The article states that Gemini 3.7 Flash executes specifications rather than inventing them, and that vague prompts therefore yield a vague game as presented in the source. The article states that the model still cannot reason. The article states that a free 27B model still writes better, as noted in the article.
Gemini 3.7 Flash was released on August 13 and was made generally available in more than 160 countries on day one. It supports up to a million input tokens and returns 64,000 output tokens, and it can read images, video, audio and PDFs while also calling tools and driving a computer. Google’s benchmark sheet places Gemini 3.7 Flash ahead of Claude Sonnet 5 and GPT-5.6 Terra on 11 of 18 tested categories, and the article lists headline metrics of 1,588 Elo on Code Arena’s web development board and 30.4% on AutomationBench.
In a zero-shot code generation test using a single prompt with no follow-ups, the model produced a playable browser game in 2 minutes and 13 seconds, with clean syntax, functioning collision and scoring logic, and visual quality noted above the price tier. The article also notes that Gemini 3.6 Flash previously could not produce a working file because its HTML was malformed, and DeepSeek found 11 bugs and shipped 8 fixes to make that earlier release playable.
These points summarize the reported release, capabilities, benchmark positioning, and test outcomes for Gemini 3.7 Flash. The article records those facts as presented in the source.
