Google Drops Three Gemini Models — Still No 3.5 Pro
Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and Flash Cyber, expanding its model lineup at the lower end of the capability spectrum. The continued absence of a Gemini 3.5 Pro raises pointed questions about where Google's frontier model development actually stands.
Original sourceGoogle announced three new additions to the Gemini family: Gemini 3.6 Flash, 3.5 Flash-Lite, and a model called Flash Cyber. All three sit in the Flash tier — Google's faster, cheaper, lower-footprint segment — rather than the Pro tier where the company competes most directly with OpenAI's o3 and Anthropic's Claude Opus. The releases expand Google's options for cost-sensitive, latency-constrained use cases, but do not advance the frontier.
The absence of Gemini 3.5 Pro is the most notable non-event here. Google has been shipping Flash variants at an accelerating pace while the Pro tier has remained static, a pattern that either reflects deliberate product sequencing or difficulty closing the gap with competitors on harder reasoning and instruction-following benchmarks. Neither Google nor its spokespeople have addressed the gap directly.
Flash Cyber appears to be a domain-specialized variant, though Google's documentation at launch is sparse on what distinguishes it architecturally from standard Flash. The naming convention itself — appending domain names to Flash — suggests Google is pursuing a segmentation strategy similar to how it once versioned Gemini by size (Nano, Pro, Ultra), now doing so by use-case instead of capability tier.
For developers already in the Google AI ecosystem, the new Flash models likely offer marginal but real improvements in throughput and cost efficiency. For anyone evaluating Google as a primary model provider against OpenAI or Anthropic, the continued Pro-tier silence will extend the wait-and-see posture that has characterized enterprise AI procurement decisions for much of 2026.
Panel Takes
The Builder
Developer Perspective
“Three more Flash variants means three more rows in my model-selection config, and I still can't answer the question my team actually has: when does Google have a Pro-tier model that competes on hard instruction-following tasks? Flash Cyber sounds interesting until you read the docs and find exactly one paragraph about what it actually does differently at inference time. Ship me a clean changelog and a benchmark with methodology, then we can talk.”
The Skeptic
Reality Check
“Google is doing something specific here: flooding the Flash tier to win on price-per-token while the Pro tier quietly stagnates, and hoping nobody notices the pattern. The direct competitor framing is OpenAI's GPT-4o mini and Anthropic's Haiku — Flash wins on GCP integration but not on raw capability, and that's fine until an enterprise buyer asks why they shouldn't just use the model that's actually best at the task. What kills this lineup in 12 months isn't a competitor — it's Google shipping a Pro model that forces everyone to admit these Flash releases were gap-fillers.”
The Founder
Business & Market
“The Flash tier is a volume play — low margin, high throughput, wins on GCP bundling discounts for customers already paying Google for infrastructure. That's a real business, but it's not a frontier AI business, and the absence of 3.5 Pro means Google is ceding the high-value, high-ASP enterprise AI contracts to OpenAI and Anthropic for another quarter. The moat here is distribution through GCP, not model quality, and that's a bet that works until one of the hyperscalers decides to stop being model-agnostic in their own marketplace.”
The Futurist
Big Picture
“The thesis embedded in this release is: capability tiers matter less than specialization tiers, and Flash Cyber is the first signal that Google is betting on domain-specific model variants rather than a single frontier model that does everything. If that bet is right — if enterprise buyers in 2027 want a model tuned for their vertical rather than the smartest general model — then this naming strategy and product architecture makes sense. The dependency is that Google has to actually ship the domain-specific performance to match the domain-specific branding, and right now the documentation doesn't support the claim.”