Google ships Gemini 3.6 Flash and two more models in a single day

While everyone was arguing about AI hitting a wall, Google quietly shipped three models in one day and confirmed the next big one is already in the oven.

Gemini 3.6 Flash: cheaper by thinking less

On July 21, 2026, Google released Gemini 3.6 Flash at $1.50 per million input tokens and $7.50 for output, down from $9.00 for output on the previous Flash.

The more interesting number is behavioral: on long agentic coding tasks, the model cuts token costs by up to 65 percent. That is not just cheaper tokens, it is fewer of them. The model thinks more economically, which compounds with the price cut on every long-running agent workflow.

The other two models

Alongside the flagship Flash came two siblings:

• Gemini 3.5 Flash-Lite at $0.30 per million input tokens and $2.50 for output, one of the cheapest capable models on the market.
• Gemini 3.5 Flash Cyber, a security-focused model that regular users cannot touch: access is limited to governments and vetted partners.

The Cyber release is a quiet signal of its own, a major lab shipping a model deliberately fenced off from the public, sold only to vetted institutional buyers.

The headline hiding in the announcement

The real news was almost a footnote: Google confirmed Gemini 3.5 Pro is on the way, and Gemini 4 pre-training is already underway.

Put the day together and the pattern is clear. Frontier intelligence is getting cheaper faster than anyone predicted, and the labs are now racing on price as hard as on IQ. Every one of these releases shifts the calculation for developers choosing a default model this summer, and the prices only seem to move in one direction. Worth tracking how fast they keep falling.