1300 166 138 Tell us what you need
News / Google

Google’s Gemini 3.5 Flash is built for autonomous agents

By Monika Tantau • Published 6 months ago
Google's Gemini 3.5 Flash is built for autonomous agents

Key takeaways

• Google has launched Gemini 3.5 Flash, a model positioned around autonomous AI agents rather than chatbots
• Google reports it runs four times faster than Claude Opus 4.7 and GPT-5.5, though independent benchmarks are not yet public
• The model ships across the Gemini app, API, Gemini Enterprise and Google AI Search, with a Pro tier to follow
• New safety mechanisms aim to curb misuse as agents take on more workflow autonomy

Most Read

Update, 21 July 2026: Google has since released Gemini 3.6 Flash, the new default Flash model, which supersedes the 3.5 Flash covered below. It is cheaper to run, uses about 17 per cent fewer output tokens on the same work and moves its knowledge cutoff forward to March 2026. See our latest Gemini launch coverage for the full breakdown. The original May report follows, kept for the record.

Google used its I/O 2026 conference to launch Gemini 3.5 Flash, and the framing is the story. The company is not selling a faster chatbot. It is positioning the model as the engine for autonomous AI agents: software that plans, acts and finishes multi-step tasks with limited human input.

Google reports the model runs four times faster than Claude Opus 4.7 and GPT-5.5, though it has not published independent benchmarks. The speed claim will get the headlines. The strategic signal matters more: the major model vendors are now competing on agents, not conversation.

What Google actually launched

Gemini 3.5 Flash is pitched first as a tool for building agents, with stronger performance on programming tasks. According to Computerworld, it is also more than twice as fast as Gemini 3.1 Pro, the version it effectively replaces for high-volume work. New safety mechanisms ship alongside it, aimed at curbing misuse as agents gain autonomy.

The model is available now through the Gemini app, the Gemini API, Gemini Enterprise and Google AI Search, with a Gemini 3.5 Flash Pro tier promised for professional users. A model wired into the API and the enterprise stack on day one is one businesses can build on now, not a preview to watch from a distance.

Why Google is leading with agents

For two years the public face of generative AI was the chatbot: a box you typed into and read answers from. Gemini 3.5 Flash reflects a different bet. As TechCrunch reported, Google is staking its next AI wave on agents rather than chatbots, systems that take an instruction, break it into steps and act.

That changes what speed is for. A fast chatbot saves a user a few seconds. A fast agent can run a longer chain of steps within a workable time, which is what makes autonomous workflows feel practical rather than sluggish. It also raises the stakes on control.

A speed claim is not a decision

Google’s four-times figure has not been independently verified, and benchmark numbers rarely map neatly onto a real workload. A neutral comparison such as our ChatGPT and Gemini breakdown can frame a shortlist, but the test that counts is your own: run a model against the tasks you actually need automated.

Control is the harder question. An agent that acts across systems needs guardrails, monitoring and an audit trail, and those are decisions about process, not model choice. Moving from chatbots to agents is a governance job, and it is worth bringing AI strategy thinking to it before a pilot, not after.

A speed claim is not a decision

What it means for Australian businesses

For Australian businesses, the takeaway is not to rush a switch to Gemini. It is that the model layer beneath your automation plans is moving fast, and a roadmap written six months ago may now assume the wrong baseline. A model built for agents changes what a realistic 2026 automation project looks like.

Two things deserve local attention. Data is the first: agentic systems touch more of your stack, so where information sits and which provider handles it becomes a Privacy Act question, not only a technical one. Dependence is the second. Google is wiring this model into its app, API, search and enterprise tools at once, which is convenient and also a lock-in decision worth making deliberately.

The models are no longer the bottleneck

Gemini 3.5 Flash will not be the last release framed this way. Anthropic, OpenAI and Google are converging on the same pitch: the value is no longer the conversation, it is the work the software completes on its own. That makes 2026 the year to get the unglamorous parts right, the integration, the guardrails and the measurement.

The faster these agents run, the more it matters that they are pointed at the right problem. That is the quieter lesson behind a speed benchmark.

Scroll to Top