Google Launches Three Gemini Models, Starts Gemini 4 Training

Google DeepMind's Triple Gemini Release

Google DeepMind announced three new Gemini models in a single release, each targeting different production AI agent needs: a stronger main model, a faster lightweight variant, and a specialized security model. The company also confirmed it has begun its "most aggressive pre-training ever" for the next-generation Gemini 4.

Gemini 3.6 Flash: Main Model with 65% Token Reduction

Gemini 3.6 Flash is positioned as the new workhorse model. Its headline improvement is dramatic token efficiency: according to Artificial Analysis Index, it uses 17% fewer output tokens than 3.5 Flash overall, and up to 65% fewer on the DeepSWE coding benchmark. It also requires fewer reasoning steps and tool calls to complete multi-step tasks, making it both faster and cheaper.

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

DeepSWE (coding): 37% → 49%

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

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

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

Demos showed 3.6 Flash orchestrating multi-agent code migration, generating 3D workflows via Gemini Canvas, and acting as an "AI interaction designer" for immersive theme studios.

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

Flash-Lite targets high-frequency, high-volume scenarios like massive document processing and agent search. It outputs 350 tokens/second—the fastest in the 3.5 series—at $0.30 per million input tokens and $2.50 per million output tokens.

Surprisingly, this lightweight model outperformed the larger Gemini 3 Flash on several 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 "brain" for task decomposition with multiple Flash-Lite instances as "workers" for parallel execution. A demo generated 25 production-ready web designs 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 cybersecurity model integrated into the CodeMender code security agent. Multiple Cyber agents collaborate to find and fix vulnerabilities, achieving state-of-the-art on the CyberGym benchmark at lower cost than larger models. The model addresses a growing gap: AI now discovers vulnerabilities faster than existing systems can patch them. Access is currently restricted; not available to general users.

Gemini 4 Pre-Training Underway

On the same day, Google confirmed it has started its "most aggressive pre-training ever" targeting Gemini 4. Following Google's typical six-month training cycle, Gemini 4 could appear by year-end. The article notes that for developers paying for tokens and building agents today, the immediate value is cost reduction: lower prices, lower token consumption, and lower per-agent-run costs. The flagship models remain "futures" — practical, affordable models are what can be deployed now.

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