At Google I/O 2026, Google announced Gemini 3.5, the new family of models aiming to combine frontier intelligence with concrete agentic capabilities. The first released model is Gemini 3.5 Flash, and the numbers it brings are significant.
4 times faster, flagship-level performance
Gemini 3.5 Flash is not just an incremental update. According to the benchmarks presented by Google, it outperforms Gemini 3.1 Pro on complex coding and agentic tasks:
- Terminal-Bench 2.1: 76.2%
- MCP Atlas: 83.6%
- CharXiv Reasoning (multimodal understanding): 84.2%
Speed? 4 times faster than comparable frontier models. Google places it in the top-right quadrant of the Artificial Analysis Index: maximum intelligence, maximum speed. An equation that until recently seemed impossible.
Intelligence that acts
The real leap of Gemini 3.5 is not in the benchmarks: it’s in the philosophy. Google explicitly talks about “frontier intelligence with action” — not just a model that responds, but one that plans, builds, iterates and solves real-world problems.
In practice, this means:
- Management of long-horizon multi-step workflows, coordinating multiple sub-agents in parallel
- Tasks that used to take a developer days, now completed in hours
- Ability to maintain context across complex sessions and advanced tool calling
Shopify uses it to analyse complex data over a long time horizon. Macquarie Bank to reason over 100+ page documents with low latency. Salesforce integrates it into Agentforce to automate multi-step enterprise tasks. Xero uses it to manage weeks-long workflows — like gathering information for 1099 tax forms — autonomously.
Gemini Spark: your personal agent running 24/7
The most surprising announcement at I/O is Gemini Spark: a personal AI agent that runs in the background around the clock, takes actions on your behalf and manages your digital life while you’re busy with other things. It’s currently rolling out to trusted testers and will come to Google AI Ultra subscribers shortly.
It’s a clear signal of direction: AI doesn’t wait for you to query it — it acts proactively, under your supervision.
What changes for AIDeskPro
Gemini 3.5 Flash is already available in the Gemini app and AI Mode in Search for billions of users. This means our users, who already use Gemini models through AIDeskPro, will directly benefit from this quality leap.
In AIDeskPro, Gemini models are available — Flash, Pro and the Lite variants — with all the guarantees we expect when working with sensitive corporate data: EU data residency (including Milan), zero data retention thanks to agreements between Open Gate and Google Cloud, and dedicated infrastructure.
Gemini 3.5 Pro is already in internal testing at Google and will arrive next month. We will follow the integration closely and, as always, release it as soon as we have verified it meets the security guarantees we expect.
The pattern repeats
Every few months, frontier models make a leap that shifts expectations. What was considered exceptional in January becomes the baseline in June. Those who adopt these tools in a structured way — in secure environments, with clear processes — accumulate an advantage that becomes harder to close with every release.
Gemini 3.5 Flash is available today. The question is not “whether to use it” — it’s “how ready are you to use it the right way”.
Original source: Gemini 3.5: frontier intelligence with action — Google Blog, 19 May 2026.