Google's New Gemini Flash Cuts Costs by 65% — But 3.5 Pro Is Still Missing

Google has released a trio of new AI models focused on speed and cost-efficiency, but the one many developers were waiting for remains absent. The new models—Gemini 3.6 Flash, 3.5 Flash-Lite, and the specialized 3.5 Flash Cyber—are designed to make AI agents cheaper to run at scale. According to VentureBeat, the flagship of this release, Gemini 3.6 Flash, can cut token costs by as much as 65% for certain long-horizon engineering tasks. This emphasis on efficiency, however, comes as the expected Gemini 3.5 Pro is still missing from the lineup.

The announcement confirms a clear tactical shift. Instead of a direct power-for-power brawl with top-tier models from competitors, Google is targeting the operational realities of deploying AI. The company is betting that for many businesses, cost and speed are more critical than having the absolute largest model available.

The Efficiency Play

The new models are engineered for specific use cases where token efficiency is paramount. As detailed by both VentureBeat and Ars Technica, the release aims to make AI agents faster and smarter without ballooning cloud computing bills. Gemini 3.6 Flash is the successor to the previous Flash model, optimized for speed and cost. Gemini 3.5 Flash-Lite is an even smaller, more lightweight version for on-device or less complex tasks, while 3.5 Flash Cyber is a specialized model fine-tuned for cybersecurity applications.

The consensus across reports is that this is a direct appeal to developers building applications that make many consecutive calls to an AI model, a process that can become prohibitively expensive. By driving down the cost per token, Google makes complex, multi-step AI agent workflows more economically viable. It is a pragmatic move, addressing a real-world bottleneck to AI adoption.

The Pro Model Question

While efficiency is a welcome improvement, the continued delay of Gemini 3.5 Pro is the central tension of this announcement. TechCrunch notes its absence raises “fresh questions” about Google’s AI strategy. The “Pro” models are Google’s direct competitors to models like OpenAI’s GPT-4o, and its non-appearance leaves a conspicuous gap in Google's portfolio.

Google has confirmed the model is still on its way. Ars Technica reports that the company says 3.5 Pro is “still in testing,” while VentureBeat notes it is “on the way.” This pattern of pre-announcing or teasing models well before they are ready is becoming a familiar part of the AI arms race. It serves to keep developers engaged and prevent them from committing fully to a competitor's platform, but it also risks eroding credibility if the delays are prolonged.

Together, these reports point to a strategy of managing expectations. Google is delivering tangible value now with cheaper models while signaling that a more powerful option is in the pipeline. This suggests Google may not be ready to compete head-to-head on flagship performance at this moment, instead choosing to compete on cost. To add another layer, Ars Technica also reports that Google is already training Gemini 4. This is a classic industry tactic: if you can't win today's battle, start talking about the next war.