- Gemini 3.6 Flash reduces output token usage by 17% versus 3.5 Flash and cuts output pricing from $9.00 to $7.50 per million tokens [1] - Gemini 3.5 Flash Cyber, a specialized cybersecurity model, found 55 confirmed issues in V8 JavaScript engine testing — outperforming Claude Opus at 36 — and is restricted to governments and trusted partners [1] - Gemini 3.5 Pro, promised at Google I/O in May for a June release, remains delayed after coding performance fell short of internal targets following a training data update
[[3]](https://9to5google.com/2026/07/16/gemini-3-5-pro-delays/) - Google teased Gemini 4 as its 'most ambitious pre-training run yet,' signaling it is already in training
[[4]](https://9to5google.com/2026/07/21/gemini-3-6-flash-launch/)
Google DeepMind on July 21 released three new AI models — Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber — broadening its mid-tier lineup while its most capable model, Gemini 3.5 Pro, remains absent more than a month past its original launch window [1] [2].
The centerpiece of the release is Gemini 3.6 Flash, which Google describes as its "workhorse model." It uses 17% fewer output tokens than the prior 3.5 Flash and cuts output pricing to $7.50 per million tokens from $9.00, while holding input pricing at $1.50 per million tokens. On the DeepSWE coding benchmark, 3.6 Flash scores 49%, up from 37% for its predecessor, and its knowledge cutoff advances from January 2025 to March 2026 [1].
The release arrives as Google faces mounting questions about Gemini 3.5 Pro, which CEO Sundar Pichai told developers at Google I/O in May would ship in June. That deadline passed without updates. Bloomberg reported that the company is working to improve the model's capabilities, "particularly in coding," after a late-June training data update produced disappointing results [3].
The New Models #
Gemini 3.6 Flash ships with a one-million-token input context window and a 64,000-token maximum output, matching its predecessor's architecture while improving efficiency. Computer use capabilities on the OSWorld-Verified benchmark rose to 83.0% from 78.4%, and performance on MLE Bench, a machine learning research benchmark, jumped to 63.9% from 49.7% [1].
Gemini 3.5 Flash-Lite targets high-throughput, low-latency workloads at a fraction of the cost: $0.30 per million input tokens and $2.50 per million output tokens. The model generates 350 output tokens per second and scores 54% on Terminal-Bench 2.1, up from 31% for the earlier Flash-Lite variant [1].
The third model, Gemini 3.5 Flash Cyber, is a purpose-built cybersecurity tool fine-tuned for vulnerability detection. In testing against Google's V8 JavaScript engine, it identified 55 confirmed issues — compared to 47 for the standard 3.5 Flash and 36 for Anthropic's Claude Opus. Access is restricted to governments and trusted partners through a program called CodeMender [1].
The 3.5 Pro Gap #
Gemini 3.5 Pro was the marquee model announcement at Google I/O in May, positioned as the company's answer to OpenAI's frontier models and Anthropic's Claude. But the promised June release window came and went without explanation [3].
According to 9to5Google, Google updated the model's training data in late June specifically to improve coding performance, but 'the results were disappointing.' Multiple outlets have reported that Google DeepMind discarded a near-ready version of the model and ordered a ground-up pre-training restart [3] [5].
Google has said only that Gemini 3.5 Pro is 'currently testing with partners' and will ship 'as soon as it's ready,' without providing a replacement timeline [2].
Pricing and Availability #
Gemini 3.6 Flash and 3.5 Flash-Lite are available immediately through the Gemini app, Google AI Studio, the Gemini API, and Android Studio. Flash-Lite will also roll out to Google Search [4].
The pricing structure positions 3.6 Flash as a direct competitor to mid-tier offerings from OpenAI and Anthropic. At $7.50 per million output tokens, it undercuts several comparable models while delivering benchmark improvements in coding and agentic tasks [1].
The cybersecurity model's government-only distribution marks an unusual access restriction for Google's AI lineup, reflecting heightened sensitivity around offensive security capabilities in AI systems [1].
Competitive Context #
The release comes as OpenAI has accelerated its own model cadence and Anthropic continues to iterate on the Claude family. The absence of Gemini 3.5 Pro leaves Google without a direct competitor to frontier-class models from its rivals at the top of the capability spectrum.
Google's decision to tease Gemini 4 — described as already in pre-training — suggests the company may be looking past the 3.5 Pro difficulties toward a next-generation architecture. Whether that signals a strategic pivot or a parallel development track remains unclear [4].
Alphabet shares traded around $342 on July 22, down roughly 1.5% on the day, though the decline was not directly attributed to the model announcements [6].
Companies mentioned #
Further sources #
[[1] Google launches Gemini 3.6 Flash and a cybersecurity model with 17% fewer outpu… ↗](https://gcn.com/google-launches-gemini-flash-cybersecurity-model/19924/)
[[2] Google releases three new Gemini models — but no 3.5 Pro — TechCrunch ↗](https://techcrunch.com/2026/07/21/google-releases-three-new-gemini-models-but-no-3-5-pro/)
[[3] Gemini 3.5 Pro delays due to coding performance, upgraded Flash model in testin… ↗](https://9to5google.com/2026/07/16/gemini-3-5-pro-delays/)
[[4] Google launches Gemini 3.6 Flash and 3.5 Flash-Lite, teases Gemini 4 — 9to5Goog… ↗](https://9to5google.com/2026/07/21/gemini-3-6-flash-launch/)
[5] Rebuilt Gemini 3.5 Pro Misses Third Deadline: Google Eyes Stopgap Release — Tec… ↗
[6] GOOGL Stock Price Quote — Morningstar ↗ The stories that matter, in one email. Free — unsubscribe anytime.