Models·3 min read·Google

Google Ships Three New Gemini Flash Models — but 3.5 Pro Is Still Missing, and It's Already Pretraining Gemini 4

On July 21 Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and a security-tuned 3.5 Flash Cyber — useful, cheap, efficient. But the flagship Gemini 3.5 Pro is still nowhere to be seen, months past its June target, and Google says it has already begun its most ambitious pretraining run yet: Gemini 4.

GOOGLE · GEMINI Three Flash models ship… …while Gemini 3.5 Pro is still nowhere to be seen 3.6 Flash −17% output tokens 3.5 Flash-Lite cheap · beats Gemini 3 3.5 Flash Cyber finds & fixes bugs Gemini 3.5 Pro promised June · still not shipped → Meanwhile, Google says it has begun its biggest pretraining run yet — Gemini 4 Shipped July 21 via the Gemini API & Gemini Enterprise · the price war grinds on
Share:

Google's answer to a rough month is to keep shipping — just not the model everyone is waiting for. On July 21 it released three new Gemini models: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a security-tuned Gemini 3.5 Flash Cyber. All useful, all in the fast-and-cheap tier. Conspicuously absent, again, is the flagship Gemini 3.5 Pro — still missing months after its June target — even as Google announced it has jumped ahead to begin pretraining Gemini 4.

The new models are real upgrades where volume actually lives. Google calls 3.6 Flash its "workhorse," saying it cuts output-token usage by 17% versus 3.5 Flash on the Artificial Analysis Index — a meaningful cost win at scale. 3.5 Flash-Lite significantly outperforms the previous Flash-Lite and even surpasses the original Gemini 3 Flash on benchmarks including SWE-Bench Pro and OSWorld-Verified, and both ship immediately through the Gemini API and Gemini Enterprise. The third, 3.5 Flash Cyber, is a lightweight model fine-tuned to find, validate, and patch software vulnerabilities, heading to governments and trusted partners through Google's CodeMender program in a limited pilot.

But the pattern is impossible to miss: Google keeps releasing Flash tiers while its top-end Pro model stays in the shop. Gemini 3.5 Pro was promised for June and has now slipped repeatedly, reportedly after engineers scrapped an earlier build over reliability gaps in recursive tool-calling and code generation. Shipping three Flash variants is exactly the "stopgap" strategy that leakers predicted — a way to stay in the game, and in the news, without the flagship that's supposed to anchor the lineup.

The Gemini 4 tease reads two ways. Optimistically, it's a flex of confidence and compute: Google says the run is its "most ambitious yet" and that it's excited by early progress. Less charitably, it looks like leapfrogging — quietly moving past a troubled 3.5 Pro toward the next generation rather than fixing the current one. Either interpretation leaves the same near-term gap: for now, the best model Google can actually give you is a Flash, not a flagship.

The strategic logic is defensible even so. The Flash tier is where the price war is fiercest and where most real API volume sits, and a 17%-cheaper workhorse plus a purpose-built security model keep Google competitive on cost and in the enterprise. But the optics are unforgiving: in the same stretch, xAI shipped Grok 4.5, OpenAI shipped GPT-5.6, and an open Chinese model, Kimi K3, topped a coding leaderboard — while Google's frontier flagship remains a promise. The Flash flurry buys time; Gemini 4 asks for patience. Whether the market grants it depends on how long "coming soon" can hold at the top of the lineup.

Want AI news before everyone else?

The morning's most important AI stories, straight to your inbox. No fluff.

Related Articles

RUMOR WATCH · UNCONFIRMED Claude Opus 5 — the evidence board Cursor · Jul 9 “Claude Honeycomb EAP” — then gone Vertex AI · ×2 listing spotted twice, vanished both times 5? the next flagship? 1M context? leaked, unverified “xhigh” mode? launch “this week”?? No official API ID · no system card · every thread on this board is still a rumor
Models

Claude Opus 5 Rumor Watch: The 'Honeycomb' Leak, a Reported 1M Context, and Talk of a Launch This Week

MOONSHOT AI · OPEN WEIGHTS Kimi K3 The largest open-weight model ever — now #1 at frontend code, ahead of Fable 5 #1 Frontend Code Arena · 1,679 pts · ahead of Fable 5 2.8T · MoE 1M context 16/896 active Weights Jul 27 Open weights on Hugging Face · native multimodal · $0.30/$3 in · $15 out per million tokens
Models

Moonshot's Kimi K3 Is the Largest Open-Weight Model Ever — and It Just Beat Fable 5 at Frontend Code

SPACEXAI · CODING & AGENTS Grok 4.5 Opus-class coding — at fast-model speed and a fraction of the cost AVG. OUTPUT TOKENS · SWE-BENCH PRO TASK Opus 4.8 67k Grok 4.5 16k 4.2× fewer tokens $2 / $6 per 1M tokens 80 TPS output speed #1 SWE Marathon Trained alongside Cursor · Terminal-Bench 2.1 83.3% · SWE-Bench Pro 64.7% · free for now in Grok Build & Cursor
Models

SpaceXAI's Grok 4.5 Bets on Efficiency Over the Crown — Opus-Class Coding at 4.2× Fewer Tokens