On July 21, Google DeepMind released three models: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber. Notably absent was Gemini 3.5 Pro — the flagship Google has been pointing at since spring. Google's product lead said it is in partner testing and they hope to land it soon, while also confirming that training has begun on a more ambitious Gemini 4.
There are two useful readings here, and only one of them is gossip.
What actually shipped, and who it's for
Gemini 3.6 Flash is the one that matters for most businesses. Google positions it as the workhorse — better at coding and knowledge work than its predecessor, and using up to 17% fewer tokens for the same job, which makes it cheaper to run. Flash-Lite is the budget tier for high-volume, low-complexity work. Flash Cyber is a security-specialized model available only to governments and trusted partners through a restricted pilot, so it is not something you will be buying.
If you already run Google Workspace, the Flash improvement is the practical news. Fewer tokens for the same output is a direct cost reduction on anything you run at volume.
The pattern worth noticing
This is the second quarter in a row where the promised top-end model slipped and the mid-tier got quietly better. It is not just Google — the industry-wide story of the past few months has been mid-tier models absorbing work that used to require a flagship. That is good news if you buy on what is available and bad news if you build a plan around a model that is still "coming soon."
How this should change your buying
Compare what is in your hands today, not what is on a roadmap. A vendor's unreleased flagship is worth exactly zero to your operation until you can run your own work through it. When you evaluate, use your real documents and your real questions on the tiers you can actually access, and let that decide.
And do not switch platforms chasing a benchmark. For most small businesses the right model is the one inside the software you already run — Google if your work lives in Workspace, Microsoft if it lives in 365 — unless a specific job is failing badly enough to justify the friction of a second tool.
The honest caveat
Model lineups change fast enough that any specific version number in this post may be stale within weeks. The durable advice is the buying discipline, not the model names. Check the vendor's current tiers before you commit to anything.
The takeaway
Google just made its cheap model better and its expensive model later. Buy accordingly: test the tier you can access this week, and treat every promised release date as a maybe.