Google released three new Gemini models on 21 July, and the interesting part is not any single one of them. It is the sequencing. The company shipped its cheapest, most token-efficient tier for running AI agents, and in the same announcement confirmed it has started its most ambitious pre-training run yet for Gemini 4.
None of these are the flagship. They sit in the Flash family, the fast and inexpensive layer built for volume, while the larger Gemini 3.5 Pro is still testing with partners and has no release date. For a business running agents at scale, the cheap tier arriving first is arguably the more useful news.
What Google released
The headline model is Gemini 3.6 Flash, now the default Flash option. It is priced at $1.50 per million input tokens and $7.50 per million output tokens, down from $9 on output for 3.5 Flash. Its knowledge cutoff moves forward from January 2025 to March 2026, and it is available now through the Gemini API, Google AI Studio, the Enterprise Agent Platform and the Gemini app.
Alongside it Google shipped Gemini 3.5 Flash-Lite, a high-throughput tier at $0.30 input and $2.50 output per million tokens, aimed at agentic search and document processing. A third model, Gemini 3.5 Flash Cyber, is a limited pilot and not something businesses can buy yet.
Why the token count matters more than the sticker price
A lower price per token is easy to understand. The more meaningful change is that 3.6 Flash uses about 17 per cent fewer output tokens than 3.5 Flash on the Artificial Analysis Index to reach the same result. When you pay per token, using fewer of them compounds with the lower rate.
It also takes fewer reasoning steps and tool calls on multi-step work, with Google citing an improvement of up to 65 per cent on the DeepSWE benchmark. For an agent that loops through dozens of steps per task, fewer steps means lower cost and faster completion, not just a cheaper unit rate.
Flash-Lite and a cybersecurity pilot
Flash-Lite is the throughput play. Google puts it at roughly 350 output tokens per second and describes it as significantly better quality than the previous 3.1 Flash-Lite. It is already rolling out inside Google Search, and its low latency suits high-volume jobs like classifying documents or handling large batches of routine queries.
Gemini 3.5 Flash Cyber is a different animal. It detects and patches software vulnerabilities inside Google’s CodeMender agent, but access is a limited pilot for governments and trusted partners, coming soon with no public pricing. Treat it as a research signal rather than a product you can adopt.
What it means for Australian businesses
For an Australian business the practical question is what it costs to run an agent all day, every day. Output tokens are where the bill accumulates, and a drop from $9 to $7.50 per million output tokens, combined with fewer tokens per task, changes the maths on whether a support agent or a document workflow pays for itself.
That said, cheaper tokens do not decide which model to build on. Model choice depends on the task, your data, and how the agent behaves under load, which is why it is worth reading a direct Claude versus Gemini comparison before committing. If you want help sizing an agent against your own workflows, our AI consulting team can model the cost per task rather than the cost per token.
The Flash tier is built for scale, so the wins show up when volume is high and the work is repetitive, not on one-off jobs.

Cheap models now, Gemini 4 teased
The sequencing is the story worth watching. Google put out its cheap, efficient agent tier while the flagship Gemini 3.5 Pro stays in partner testing with no date, and while it says broad availability will come as soon as it is ready.
At the same time it confirmed it has started its most ambitious pre-training run yet for Gemini 4. That is a pre-training run, not a release, and it is likely months away. The shape is clear though: the affordable models to run agents today, the big one held in reserve. For businesses, the sensible move is to build on what ships now and stay ready to swap.
