Google finally launched today. Just not the model anyone was waiting for.

This morning Google DeepMind shipped three new Gemini models at once: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a gated Gemini 3.5 Flash Cyber. All three are Flash-tier, cheaper, faster, efficiency-focused. And conspicuously absent, again, was Gemini 3.5 Pro, the flagship Sundar Pichai promised back in May and has now delayed three separate times.

The subtext is hard to miss. Google is shipping stopgaps to stay in the conversation while its top-tier model stays stuck in testing, and this week the scoreboard finally caught up with it. Here is what launched, and why the thing that did not launch is the real story.

Gemini 3.6 Flash is Google's new workhorse, and its pitch is efficiency, not raw power. It uses about 17 percent fewer output tokens than 3.5 Flash on the same tasks, takes fewer reasoning steps and tool calls in multi-step workflows, and generates cleaner code with fewer unwanted edits. Google also cut the price to $1.50 per million input tokens and $7.50 per million output, down from $9 on output, with a 1-million-token context window. It is generally available today across the Gemini API, AI Studio, Vertex AI, Google's Antigravity agent platform, and the Gemini app, and it is rolling into GitHub Copilot.

Keep expectations calibrated, though. This is a mid-tier model, and it is being judged like one: builders welcomed the price and efficiency, but it does not top any leaderboard, and some early reviewers called it underwhelming next to rivals like Sonnet 5, Grok 4.5, and GPT-5.6's cheaper tiers. If you run high-volume agentic or coding workloads on Gemini, 3.6 Flash is a genuine cost-and-efficiency upgrade worth testing. If you were waiting for something to challenge the frontier, this is not it.

The supporting cast is more interesting than it looks. Gemini 3.5 Flash-Lite is the budget workhorse, priced at just $0.30 per million input and $2.50 output, running around 350 tokens per second for high-volume jobs like search, translation, and document processing. It actually beats Google's older 3 Flash on several evals, including SWE-Bench Pro at 54.2 percent versus 49.6, and it exposes configurable thinking levels so you can dial cost against depth.

The third release is the one to watch. Gemini 3.5 Flash Cyber is fine-tuned to find, validate, and patch software vulnerabilities, and it runs many cheap instances in parallel inside CodeMender, Google's code-security agent. But Google is keeping it locked, available only to governments and trusted partners in a limited pilot. That gating instantly reopened the dual-use debate, because an AI built to find exploits is exactly as useful to an attacker as a defender, and Google is deciding who gets to hold it.

🚧 The Missing Flagship

Google shipped three models today and still could not ship Gemini 3.5 Pro. Pichai promised it "next month" at I/O in May, June came and went, a July 17 target came and went, and it is now on its third slip, reportedly stuck after engineers found structural failures in the rebuilt model. Bloomberg reported the delay has frustrated Google's own engineers and researchers, who worry the company is losing its edge to Anthropic and OpenAI.

The scoreboard agrees. This week Google fell out of the top five AI labs on the independent Artificial Analysis Intelligence Index for the first time, with Anthropic, OpenAI, SpaceXAI, Meta, and a Chinese open-source lab all ranking above it, in a stretch where Grok 4.5, three GPT-5.6 variants, Muse Spark 1.1, and Kimi K3 all shipped. Google's answer is to keep releasing efficient Flash models to hold the line, and it confirmed today that it has already started pre-training Gemini 4. The bet is clear: skip past the troubled 3.5 Pro and win the next round. Whether the market waits that long is the open question.

The stories moving fast beyond today's headlines:

Google DeepMind launched three new Gemini models today, Gemini 3.6 Flash, 3.5 Flash-Lite, and a gated 3.5 Flash Cyber, and skipped the one everyone actually wanted. Gemini 3.6 Flash is the headline, a cheaper, more efficient workhorse using 17 percent fewer output tokens than 3.5 Flash at $1.50 and $7.50 per million with a 1-million-token context, available now across the Gemini API, AI Studio, Vertex, Antigravity, and rolling into GitHub Copilot, though reviewers judged it a solid mid-tier upgrade rather than a frontier contender. Flash-Lite goes ultra-cheap at $0.30 and $2.50 for high-volume work, and Flash Cyber, a vulnerability-finding model, is gated to governments and trusted partners, reopening the dual-use debate. But the real story is the flagship that did not ship: Gemini 3.5 Pro is now on its third missed deadline, reportedly stuck after a failed rebuild, and this week Google fell out of the top five AI labs on the Artificial Analysis index for the first time, while Grok 4.5, three GPT-5.6 variants, Muse Spark 1.1, and Kimi K3 all launched around it. Google's move is to keep shipping Flash stopgaps and skip ahead, confirming today that Gemini 4 pre-training has begun. For builders, 3.6 Flash is worth testing for cheap agentic and coding workloads, but the frontier question stays open until Pro, or Gemini 4, actually ships.

Stay building. 🤖