Google releases Gemini 3.5 Flash, a new model built for agentic tasks

Google has added Gemini 3.5 Flash to its lineup, the first model in the Gemini 3.5 family, which the company describes as combining "frontier intelligence" with real-world agentic action. The model rolled out simultaneously in the Gemini app, AI Mode in Google Search, the Google Antigravity development platform, the Gemini API, and enterprise deployments through Gemini Enterprise.

What the new model brings

According to Google, Gemini 3.5 Flash outperforms the family's previous leader, Gemini 3.1 Pro, on coding and agentic tasks, even though it is formally the "fast" variant rather than the flagship "Pro" model. The company says the model handles long, multi-step tasks that require planning and executing successive steps without human supervision at every stage.

In benchmarks published by Google, the model scored 76.2 percent on Terminal-Bench 2.1, a test of the ability to work independently in a terminal, and 83.6 percent on MCP Atlas, an evaluation of tasks that use the Model Context Protocol to connect the model with external tools. On GDPval-AA, a test comparing the model's work quality against that of human professionals, Gemini 3.5 Flash scored 1,656 Elo points.

Speed as a selling point

Google is heavily emphasizing the model's performance, claiming it generates output tokens four times faster than other leading models on the market. On the Artificial Analysis Intelligence Index, which plots the trade-off between speed and model intelligence, Gemini 3.5 Flash landed in the upper right quadrant alongside the market's most efficient solutions.

Gemini 3.5 is built to help carry out complex, agentic workflows - Google team

Gemini 3.5 Pro still on the way

Flash is, for now, the only variant of the 3.5 family that has been released. Google announced that a more powerful Gemini 3.5 Pro, already being tested internally, will reach users next month. That announcement came before a series of delays that tech media later reported on: Google first pushed the Pro launch to July 17, citing a ground-up rebuild of the model after discovering issues with recursive tool calling and SVG graphics generation.

Splitting the launch into two stages, first the cheaper, faster Flash, then the flagship Pro, follows the strategy Google used with previous model generations. This lets the company test new capabilities on a wider group of users and gather feedback before releasing the most computationally demanding variant.

Impact on developers and businesses

For developers, the key point is that the model is available in Google Antigravity and in the Gemini API for Google AI Studio and Android Studio, making it possible to build it directly into development tools and coding agents right away. Among early adopters, Google lists companies from a range of industries, from retail (Shopify) to banking (Macquarie Bank) to data management (Databricks), suggesting the model is aimed primarily at competing in business and developer use cases rather than just consumer chat.

The Gemini 3.5 Flash launch arrives amid intensifying competition in the large language model market, where OpenAI is developing its GPT-5.6 family and Chinese companies such as Moonshot AI and Zhipu are releasing increasingly capable open models. For Polish companies and developers, this means another option to test in tools that already rely on Google's API, though pricing for the new variant has not yet been officially confirmed.