Google DeepMind released three new proprietary AI models on July 21, 2026, and the headline is not a new frontier flagship. It is efficiency. Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a restricted Gemini 3.5 Flash Cyber variant are designed to make AI agents faster, smarter, and cheaper at scale.
The economics of running autonomous software agents in production come down to a fixed equation few vendors advertise directly. A model needs to reason through a multi-step task competently, but every extra token it generates adds cost and delay to a workflow that might run thousands of times an hour. Google's answer splits that trade-off across three models.
Gemini 3.6 Flash: the new workhorse
Gemini 3.6 Flash is priced at 7.50 per million output tokens. Output pricing drops from the prior model's $9.00. The bigger savings may come from token efficiency. On the Artificial Analysis Index, 3.6 Flash uses 17% fewer output tokens than 3.5 Flash. On the Datacurve DeepSWE benchmark, Google reports drops in token usage of up to 65%.
Quality gains accompany the efficiency. On DeepSWE, 3.6 Flash scores 49% versus 37% for its predecessor. On MLE Bench, it reaches 63.9% versus 49.7%. On OSWorld-Verified, it hits 83.0% versus 78.4%. On GDPval-AA v2, a knowledge-work benchmark, it scores 1421 versus 1349. Google says coding output has fewer unwanted edits and reduced execution loops, while multi-step workflows take fewer reasoning steps and tool calls.
The model ships with a built-in client-side computer-use tool through the Gemini API and Gemini Enterprise platforms, removing the custom intermediary software engineers previously built. It also carries enhanced Frontier Safety safeguards covering chemical, biological, radiological, and nuclear misuse, plus cyber-offense misuse.
Early customers are already integrating it. Figma has folded 3.6 Flash into its prototyping infrastructure. Legal technology platform Harvey and research tool Hebbia route data through it for multimodal document work: ingesting raw financial filings, parsing document structure, reading embedded charts, and producing draft reports.
Gemini 3.5 Flash-Lite: the throughput play
Flash-Lite targets a different job. At 2.50 per million output tokens, it is cheap enough that engineering teams can route simple, high-volume subagent requests to a minimal thinking level and reserve higher thinking levels for multi-step work. Google reports 350 output tokens per second, the fastest in the 3.5 series.
On the GDM-MRCR v2 long-context test, Flash-Lite records 72.2% versus 60.1% for its predecessor. Its GDPval-AA v2 score nearly doubled, from 642 to 1140. Google also reports that Flash-Lite beats the larger Gemini 3 Flash on SWE-Bench Pro, 54.2% versus 49.6%, and on OSWorld-Verified, 74.0% versus 65.1%.
Flash Cyber: the restricted specialist
The third model, Gemini 3.5 Flash Cyber, is built for vulnerability discovery, validation, and patching inside Google's CodeMender system. It does not have public API pricing or self-serve access. Google is making it available only to governments and trusted partners through a limited pilot, acknowledging the dual-use nature of cybersecurity AI.
The competitive landscape
At 10.50 and Gemini 3.1 Pro Preview at 0.42 total, MiMo-V2.5 Flash at 1.75. The value proposition is not being the cheapest. It is being the most token-efficient for agentic workloads that generate long reasoning chains.
Google says Gemini 3.5 Pro remains in partner testing ahead of a full release, and pre-training for the next Gemini 4 architecture is already underway. Until then, developers must optimize their systems using the Flash architecture. The right next step, as always with model migrations, is not to replace every endpoint at once. Choose representative tasks, record total tokens and tool calls, measure retries and human corrections, and migrate only when the new model improves the cost of a completed, reviewed task.