Version 3

Google released Gemini 3.6 Flash and 3.5 Flash-Lite on July 21, 2026, replacing 3.5 Flash just two months after its Google I/O debut, while a cybersecurity-specialized variant (3.5 Flash Cyber) remains restricted to governments and trusted partners due to its ability to autonomously find and exploit production vulnerabilities [1][4]. Gemini 3.5 Pro has still not shipped, held back by coding performance shortfalls [7]. Google disclosed it has begun pre-training Gemini 4, with no timeline attached [1]. Commentary is now framing the three-model lineup as a deliberate move away from a single general-purpose flagship toward purpose-built specialization [2].

Google is cycling Flash models faster than standard product timelines while a Pro-tier model sits unreleased and a security-specialized model is kept behind access controls rather than capability limits. The gatekeeping argument for Flash Cyber — that restricting access is the appropriate safeguard for a model that generates working exploits — faces an obvious durability question: it holds only as long as no competitor ships a comparable model publicly [2].

Google released three Gemini models in quick succession in July 2026. Gemini 3.6 Flash, released July 21, replaces 3.5 Flash just two months after its Google I/O debut, claiming 17% fewer output tokens and improvements on benchmarks where 3.5 Flash had underperformed, including DeepSWE coding (49% vs. 37%) and OSWorld computer use (83.0% vs. 78.4%) [1]. Gemini 3.5 Flash-Lite, released the same day, targets high-volume latency-sensitive workloads at 350 output tokens per second and is being integrated into Google Search [1][2]. 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens [2]. Ars Technica noted the unusual pace of deprecating a two-month-old flagship and argued that Google's focus on token efficiency for enterprise cost reduction may have contributed to the coding shortfall that 3.6 Flash was built to address [3].

Four days before the Flash releases, Google announced Gemini 3.5 Flash Cyber, trained specifically for offensive and defensive security work. In internal evaluations it found 55 unique confirmed vulnerabilities in Chrome's V8 JavaScript engine, versus 47 by mainline 3.5 Flash and 36 by Claude Opus 4.6 [4]. In a two-hour test on Google's own production systems, it uncovered remote code execution vulnerabilities in public APIs and a memory-corruption flaw in a sensitive service, then generated a functional RCE exploit bypassing both ASLR and W^X mitigations [4]. Google restricted the model to governments and trusted partners via its CodeMender platform, treating access controls rather than capability reduction as the appropriate safeguard. The Verge and CyberNews framed the model primarily as a cheaper alternative to large dedicated AI security models such as Mythos [5][6]. The Neuron added a sharper point: that the gatekeeping logic holds only as long as no competitor ships an equivalent model without restrictions [2].

Running alongside the Flash releases is the unresolved delay of Gemini 3.5 Pro, which had been expected in June. Bloomberg reported Google held the model back because coding performance fell short of internal targets [7], and Alphabet's share price fell on that report [8]. Multiple outlets published analysis on the delay in late July [9][10][11][12], though Google's public communications have offered no specifics. Some social media commentary has connected Google's disclosure that Gemini 4 pre-training has begun to the view that 3.5 Pro development proved disappointing [13], though Google has not said so. Google described the Gemini 4 run as its most ambitious pre-training effort yet, without attaching a timeline or capability claims [1].

Google's three-model lineup has drawn a framing argument: that it reflects a deliberate departure from single-flagship thinking toward purpose-built specialization. The Neuron argued that Google's approach positions agents as working best when a model is built for the job rather than maximally generalist, and described the lineup as Google quietly acknowledging that the 'one model to rule them all' era is ending [2]. Whether that framing reflects a stable product strategy or an artifact of the 3.5 Pro delay remains unresolved.