Google Ships Three Gemini Models at Once — and Says Gemini 4 Pre-Training Has Started

Google released three Gemini models on July 21 — and the one everyone was waiting for was not among them. Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and a security-tuned Gemini 3.5 Flash Cyber all shipped together, while the flagship Gemini 3.5 Pro remains in partner testing with no date attached.

Cheaper output, fresher knowledge

Gemini 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output, down from $9 on the previous Flash. Google says it emits 17% fewer output tokens on the Artificial Analysis index and up to 65% fewer on the DeepSWE agentic-engineering benchmark — the number that matters most for agent operating costs. Its knowledge cutoff moves from January 2025 to March 2026, across the full 1M-token context window.

The benchmark jumps

Against 3.5 Flash, the new model scores 49% versus 37% on DeepSWE, 63.9% versus 49.7% on MLE Bench and 83.0% versus 78.4% on OSWorld-Verified. Gemini 3.5 Flash-Lite, at $0.30/$2.50 per million tokens and about 350 output tokens per second, nearly doubles its predecessor on Terminal-Bench 2.1 (54% versus 31%) and is rolling into Google Search.

A model most people cannot use

Flash Cyber is the outlier: tuned to find and fix security vulnerabilities, it ships as a limited-access pilot inside CodeMender, restricted to governments and trusted partners on dual-use grounds, with no public API and no published price. Google claims competitive frontier performance on the CyberGym benchmark.

And then, Gemini 4

The line buried in the announcement carried the most weight: Google says it has begun "our most ambitious pre-training run yet, for Gemini 4." On 3.5 Pro, the company would only repeat that it is testing with partners and will ship "as soon as it's ready" — after telling developers in May it was a month away.