The wait for Gemini 3.5 Pro is over. Mostly.
Google isn’t launching it. Not today, not yet.
Instead, the company dropped two new entries into its Flash lineup. They also introduced a specialized tool for cybersecurity. It is a strange pivot. You would expect a pro-tier release. You’d expect the heavy lifter. You don’t get it.
On July 21, Google detailed the new models. They call them Gemini 3.6 Flash. They also announced Gemini 3.5 Flash-Lite. Both are built for efficiency. Low latency. Reliability at scale. This matters if you are running AI agents. Not just chatting. Actually working.
There is a third model. It is narrower. Gemini 3.5 Flash Cyber. It works with Google’s CodeMender security agent. It detects vulnerabilities. It patches them.
Where is Gemini 3.5 Pro?
You remember Sundar Pichai’s promise. He said Gemini 3.5 Pro was coming in June. This was at Google I/O back in May. He was confident.
The timeline has shifted. General availability is still “as soon as it’s ready.” That is code for “later.”
Right now, the model is testing with partners. Only select ones. Not the public.
Why the delay? The AI market is moving fast.
Anthropic released Fable 5. OpenAI launched GPT-5.6 Sol. Meanwhile, Chinese lab Moonshot dropped Kimi K3. It is affordable. It is open-source. It is competitive. Google cannot afford to be left behind. They need to fill the gap with what they have.
They have Flash models. They have a security tool. They are rolling those out now.
The Efficiency Play: Gemini 3.6 Flash
This is the new general-purpose workhorse.
Gemini 3.6 Flash improves on 3.5 Flash. Better coding. Better knowledge work. Better multimodal tasks.
Google is pricing it aggressively.
Input tokens cost $1.50 per million. Output tokens are $7.50 per million. That is lower than the previous version. Cheaper. Faster.
Benchmarks show gains. They didn’t just guess. They tested. The results are there.
The Speed Demon: Gemini 3.5 Flash-Lite
Speed is everything for some use cases.
Google claims this is the fastest in the 3.5 series.
Artificial Analysis figures cited in the post support this. It hits 350 output tokens per second. That is fast.
The price is even lower than 3.6 Flash.
Input: $0.30 per million tokens. Output: $2.50 per million tokens.
Who needs this? High-throughput workloads. Latency-sensitive tasks. Agentic search. Document processing. It outperforms the prior Flash-Lite on agentic benchmarks.
It is not for everyone. It is for scale.
The Secret Weapon: Gemini 3.5 Flash Cyber
This one is different.
It is designed for one thing: detecting and patching software vulnerabilities.
It is deployed inside CodeMender. Google’s code security agent.
You cannot just sign up and use it. It is not broadly available.
Google is restricting access. They are offering it exclusively to governments and selected partners. A limited-access pilot. Why? Dual-use risks. A model that can find security flaws is also a weapon. They are cautious.
“We are prioritizing safety by limiting access to this specific capability.”
That is the implication. You have to earn it. You have to be a partner. Or a government.
Safety and Scope
Safety is a big theme here.
Gemini 3.6 Flash includes expanded measures. It resists jailbreak attempts. Specifically those involving chemical, biological, radiological, or nuclear misuse. It also targets cyber offenses.
Google aims to avoid unnecessary refusals for legitimate uses. Balance is hard. They are trying.
What’s Next?
Training for Gemini 4 has begun.
That is all we know. For now.
The release of Gemini 3.6 and 3.5 Flash-Lite feels like a stopgap. Or a strategic shift. Maybe both.
They are betting on efficiency. On speed. On specific verticals like security.
The Pro model is still in the oven. It is testing with partners.
When it drops? No one says.
The market doesn’t stop. Anthropic keeps pushing. OpenAI keeps pushing. Google is responding. But with Flash models first.
Interesting choice. You can argue about the strategy. But the benchmarks show improvements. The prices are competitive. The security tool is unique.
It remains to be seen if this fills the void left by the delayed Pro release. For now, the Flash models are here. The rest is silence.
