Google Unveils Three Gemini Models, Starts 'Most Aggressive' Gemini 4 Training

Google DeepMind's Triple Release and Gemini 4 Announcement

Google DeepMind announced three new Gemini models in a single day: Gemini 3.6 Flash , Gemini 3.5 Flash-Lite , and Gemini 3.5 Flash Cyber . Simultaneously, Google confirmed it has begun the "most aggressive pre-training in its history" targeting the next-generation Gemini 4 , with a possible release by year-end based on a typical six-month training cycle.

Gemini 3.6 Flash: Token Efficiency and Performance Gains

Positioned as the flagship workhorse, Gemini 3.6 Flash emphasizes dramatic token savings. According to Artificial Analysis Index, it uses 17% fewer output tokens than its predecessor 3.5 Flash. On the DeepSWE coding benchmark, token consumption drops by up to 65% . The model also requires fewer reasoning steps and tool calls, translating to faster and cheaper execution.

Pricing is reduced to $1.50 per million input tokens and $7.50 per million output tokens , cheaper than 3.5 Flash. Despite lower cost, benchmark scores improve significantly:

DeepSWE (coding): 37% → 49%

MLE Bench (ML research): 49.7% → 63.9%

OSWorld-Verified (computer control): 78.4% → 83%

GDPVal-AA v2 (knowledge work): 1,421 tasks completed , 70+ more than previous generation

Demos showcase multi-agent orchestration for complex code migration, 3D workflow creation via Gemini Canvas (producing a photography-grade texture extractor), and immersive theme studio design as an "AI interaction designer."

Gemini 3.5 Flash-Lite: Speed and Cost for High-Volume Workloads

Flash-Lite targets high-throughput scenarios like massive document processing and agent search. It delivers 350 tokens/second output speed—the fastest in the 3.5 series—at a price of $0.30 per million input tokens and $2.50 per million output tokens .

Remarkably, this lightweight model outperforms the larger Gemini 3 Flash on key benchmarks:

SWE-Bench Pro: 54.2% vs 49.6%

OSWorld-Verified: 74.0% vs 65.1%

The suggested architecture pairs 3.6 Flash as the "planner" decomposing tasks, with multiple Flash-Lite instances as "workers" executing in parallel. An official demo generated 25 production-ready web design proposals almost instantly. Flash-Lite is already live in the Gemini App and rolling into Google Search.

Gemini 3.5 Flash Cyber: Specialized Vulnerability Detection

Flash Cyber is a dedicated security model for finding and fixing vulnerabilities. Integrated into the CodeMender code security agent, multiple Cyber agents collaborate to achieve state-of-the-art results on the CyberGym benchmark at lower cost than larger models. The model addresses the growing gap where AI discovers vulnerabilities faster than existing systems can patch them. However, Flash Cyber is not publicly accessible at this time.

Gemini 4 Pre-Training Underway

Google stated that internal pre-training for Gemini 4 has started, calling it the "most aggressive pre-training ever." Following the historical six-month cadence, a release could arrive as early as late 2024. The article notes that while flagship models remain "futures," today's releases deliver immediate cost reductions—lower token usage, lower prices, and lower per-agent operational costs—for developers building production AI agents.

Reference: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/?utm_source=tw&utm_medium=social&utm_campaign=og