Google cuts AI token costs with new Gemini models
Gemini 3.6 Flash delivers up to 65% savings on long-horizon engineering tasks.
1New Gemini Models Announced
Google DeepMind unveiled three new token-efficient AI models: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. These models aim to make AI agents faster, smarter, and more cost-effective. They promise significant savings, especially for long-term engineering tasks.
2Pricing Details
Gemini 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens. The 3.5 Flash-Lite model offers an even more affordable option at $0.30/$2.50. Google's previous generation 3.1 Flash-Lite remains the most economical choice currently available.
3Industry Competition
Google's new models compete in the mid-to-high segment against rivals like Xiaomi's MiMo-V2.5 Flash and DeepSeek's deepseek-v4-flash. However, their token efficiency provides cost advantages, especially noticeable in extended workflows where savings compound significantly.
4Specialty Use: Flash Cyber
The cybersecurity-focused Gemini 3.5 Flash Cyber model's pricing remains undisclosed. Designed for bug patching and vulnerability detection, this specialized model will be available exclusively to governments and trusted partners through the CodeMender platform.
5Token Efficiency
Gemini 3.6 Flash uses 17% fewer output tokens than its predecessor. In multi-step engineering benchmarks like DeepSWE, savings reach up to 65%. The model completes identical workflows with fewer reasoning steps and tool calls, operating like a fuel-efficient hybrid vehicle.
6Technical Specifications
The new models feature a 1-million-token input context window with a 64,000-token output limit. While Google hasn't detailed architectural changes, they emphasize the models' reduced verbosity and faster response times as key efficiency drivers.
7Availability
Gemini 3.6 Flash and 3.5 Flash-Lite are now available through Google AI Studio, Android Studio, and consumer apps. The specialized Flash Cyber model will launch soon on CodeMender. All models remain proprietary, accessible only through Google's official APIs.
8The Awaited Flagship: Gemini 3.5 Pro
Google's promised flagship model, 3.5 Pro, remains unreleased. The company states it's currently being tested with partners. This delay has allowed competitors like OpenAI and Anthropic to release multiple generations of more powerful models.
9The Future of AI Agents
Google's move signals that AI's future lies in autonomous agents. Experts compare older models to fuel-guzzling trucks, while the Flash series represents nimble hybrid couriers. This transition could significantly reduce enterprise costs for AI implementation.
10Industry Impact
These token efficiency breakthroughs hold particular importance for companies running long-term software engineering and automation projects. Google's advancements may pressure competitors to follow suit, potentially benefiting the entire AI ecosystem through improved cost structures.
Original article: VentureBeat — Google, yeni AI modelleriyle token maliyetlerini düşürüyor