Gemini 3.5 is out! Google boasted brutal speed and a new agent, Gemini Spark

In addition to new smart glasses and the generative model Gemini Omni, this year’s Google I/O 2026 conference brought another significant innovation – the next generation of Gemini 3.5 models. This time, Google chose an unconventional approach. It is first releasing the Flash variant, which, according to its own benchmarks, is designed to match or surpass the more powerful 3.1 Pro model. The blurring of lines between fast and flagship models has been a clear trend across the entire AI industry recently.

Flash that performs like Pro

Flash series models in Gemini have so far been designed to offer reasonable performance at a reasonable price, and most importantly, speed. However, the new generation 3.5 Flash is intended to offer the best of both worlds. Google states that the new Flash model surpasses the current flagship Gemini 3.1 Pro in coding and so-called agentic tasks. Specifically, it achieves 76.2% in the Terminal-Bench 2.1 test (performing tasks in the terminal), 83.6% in MCP Atlas (working with tools), and 84.2% in CharXiv Reasoning (understanding multimodal inputs).

At the same time, it maintains speed – according to Google, it generates output up to four times faster than competing top models (the company does not specify which ones, but GPT and Claude from OpenAI and Anthropic can be assumed). This is important for long tasks where every second multiplies. Banks and fintech companies, according to Google, are already using 3.5 Flash to automate processes that previously took several weeks.

Personal Agent Gemini Spark

Google also demonstrated the agentic aspect of the new model in practice. Along with 3.5 Flash, it introduced Gemini Spark – a personal AI agent designed to run in the background 24 hours a day and, under user supervision, perform tasks in their digital life. Examples include preparing financial documents, organizing unstructured files by content, or automating multi-step workflows.

Gemini Spark is being rolled out to selected testers starting today, and next week, subscribers to the Google AI Ultra plan in the USA will receive it as a beta version. Google has not announced a global launch date, nor has it yet commented on when Spark will arrive in Europe.

Security and Where to Find the Model

With every newer AI model, Google also addresses protection against its misuse. Gemini 3.5 was developed in accordance with the company’s Frontier Safety Framework, with particular emphasis on strengthening mechanisms against generating content related to cyber threats and CBRN risks (chemical, biological, radiological, and nuclear weapons). Google also claims that the model less frequently incorrectly rejects legitimate queries – which was a common problem with older versions.

Gemini 3.5 Flash is globally available starting today in the Gemini app, in AI Mode in Google Search, and for developers via Google AI Studio, Android Studio, and the Antigravity platform. For enterprise customers, it’s available within Gemini Enterprise. Google has not yet specified pricing for paid API tiers. The more powerful Gemini 3.5 Pro, which the company states it is already using internally, is expected to arrive in June.

Do you use Gemini daily?

Source: Google Blog