Google ships Gemini 3.6 Flash, skips 3.5 Pro
Three new Gemini models land with big efficiency gains on coding and agentic benchmarks, but the Pro tier everyone expected did not ship this round.
Google released three Gemini models on July 21: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a security-tuned Gemini 3.5 Flash Cyber, all pitched on efficiency rather than a new capability ceiling. Google says 3.6 Flash cuts output token usage 17% versus 3.5 Flash on the Artificial Analysis Index while improving up to 65% on the DeepSWE code-editing benchmark and scoring 83.0% on OSWorld-Verified computer-use tasks, versus 78.4% for its predecessor. It prices at $1.50 per million input tokens and $7.50 per million output tokens.
Gemini 3.5 Flash-Lite is the bigger jump for agentic workloads specifically: Google cites 350 output tokens per second, a Terminal-Bench 2.1 score of 54% against 31% for the prior Flash-Lite, and a SWE-Bench Pro result (54.2%) that beats the full 3 Flash model (49.6%). It prices at $0.30 per million input tokens and $2.50 per million output tokens. Both models are live now in the Gemini API, Google AI Studio, Android Studio, and the Gemini Enterprise Agent Platform, with 3.5 Flash-Lite also rolling into Google Search. Gemini 3.5 Flash Cyber, built into Google’s CodeMender security-agent work, stays a limited pilot for governments and other trusted partners rather than a public release.
Update, July 23: Buried in the same announcement, Google confirmed for the first time that a Gemini 4 pre-training run is underway:
We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress.
It’s the first time Google has used the Gemini 4 name in its own words, but the company attached no timeline, no capability claims, and no access details. Google typically runs roughly six-month training cycles, which would put a plausible Gemini 4 release toward the end of 2026 if that cadence holds, though pre-training starting is not the same as a release date.
What’s notably absent is a Gemini 3.5 Pro. TechCrunch’s framing, that Google shipped speed and cost gains while its top-tier model keeps slipping, matches a pattern operators are already routing around: when frontier labs stagger releases, the cheap, fast tier often lands well ahead of the flagship, and workloads that don’t strictly need frontier reasoning have a new cheaper default. If your agents lean on tool use or long-horizon computer control, 3.6 Flash and 3.5 Flash-Lite are worth a head-to-head test against whatever you’re routing to today, since the benchmark gaps here are large enough to matter on cost alone, not just capability. It’s the same trade playing out with Cursor’s new request-level model router: a vendor’s own benchmark picks tell you where they want the spotlight, and pricing tells you where they expect volume.