Skip to main content
FeaturedDaily
Back to all news

Google

Gemini 3.5 Pro Still in Preview; GA Reportedly Set for 17 July

Google rebuilt Gemini 3.5 Pro from scratch, delaying it into July. Leaked reporting now points to a 17 July launch — and a security-benchmark twist.

By , Editor-in-Chief · FeaturedDailyVerified July 2026

The answer

Gemini 3.5 Pro is still in preview; leaks put general availability on 17 July after a full rebuild.

What happened

While OpenAI and Anthropic sprinted out of a July government review, Gemini slipped. Teased at I/O in May, it missed a June target and a 30 June GA date, and as of 9-10 July remained in limited Vertex AI enterprise preview. Leaked details reported around 10 July now place general availability on 17 July 2026 — a date Google has not confirmed.

The reason is now understood: DeepMind reportedly abandoned the original 2.5 Pro base and ran an entirely new pretraining run, effectively rebuilding the model from scratch mid-cycle. Restarting a frontier model's pretraining is one of the most expensive moves a lab can make — it burns months of compute and pushes every downstream date back. That is a costly, revealing bet, and the simplest explanation for why the launch kept slipping past target after target.

On July 9, with both GPT-5.6 and Grok 4.5 launching publicly, Gemini 3.5 Pro enters its most difficult competitive context.

Source: BuildFastWithAI · 9 July 2026

The specs and the security-threshold irony

Leaked specs cite a 2M-token context window (reportedly the largest of any production frontier model), Deep Think reasoning gated to the $250/mo Ultra tier, and pricing near $1.25 input / $10 output per 1M tokens. Treat all of it as leaked, not official — no published benchmarks or confirmed pricing as of 10 July.

Here is the twist. Reporting framed Gemini 3.5 Pro as the only major frontier model set to ship without US government restrictions — apparently because its offensive-security scores sat below the unofficial threshold that triggered scrutiny for OpenAI's GPT-5.6 and Anthropic's Fable/Mythos. So Google's slower, less cyber-capable model is the one that ships freely. "Below the threshold" is an inference from reporting, not a Google statement.

What's next

Watch 17 July for a possible GA — and for Google to publish official specs, benchmarks and pricing that either confirm or deflate the leaks. If the 2M-token window and Deep Think land as reported, Gemini gets a real enterprise pitch; if the date slips again, the rebuild story only gets louder. Until then, Gemini 3.5 Pro stays a preview-only product carrying a big-context, high-tier reasoning promise and an awkward compliment: it ships freely because it scored less dangerous.

Frequently asked questions

Is Gemini 3.5 Pro available yet?
No. As of 9-10 July 2026 it remained in limited Vertex AI enterprise preview, having missed a 30 June GA date. Leaked reporting points to a 17 July launch, which Google has not confirmed.
Why was it delayed?
DeepMind reportedly abandoned the 2.5 Pro base and ran an entirely new pretraining run — rebuilding the model from scratch mid-cycle, which pushed the timeline back.
Are the specs official?
No. The 2M-token context window, Deep Think on the $250/mo Ultra tier, and roughly $1.25 input / $10 output pricing all come from leaks. There were no published official benchmarks or confirmed pricing as of 10 July.
Why does it ship without US government restrictions?
Reporting says its offensive-security scores fell below the unofficial threshold that triggered scrutiny for OpenAI and Anthropic. That inference isn't a Google statement.

Sources

← All news