Google Launches Three Gemini Models and Starts Gemini 4 Training

Google DeepMind Unveils Three New Gemini Models

On the same day, Google DeepMind announced three new Gemini models targeting production AI agents: Gemini 3.6 Flash , Gemini 3.5 Flash-Lite , and Gemini 3.5 Flash Cyber . Together they aim to make agents faster, smarter, and cheaper.

Gemini 3.6 Flash: 65% Token Reduction on Coding Benchmarks

Positioned as the flagship of the trio, Gemini 3.6 Flash emphasizes token efficiency. According to Artificial Analysis Index, it uses 17% fewer output tokens than its predecessor 3.5 Flash. On the DeepSWE coding benchmark, token savings reach 65% . The model also requires fewer reasoning steps and tool calls, resulting in lower latency and cost.

Pricing is reduced to $1.50 per million input tokens and $7.50 per million output tokens , cheaper than 3.5 Flash.

Benchmark improvements over 3.5 Flash:

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

Gemini 3.5 Flash-Lite: Fastest 3.5 Series, Beats Larger 3 Flash

Flash-Lite targets high-throughput, cost-sensitive workloads. It delivers 350 tokens/second output speed — the fastest in the 3.5 family — at a price of $0.30 per million input tokens and $2.50 per million output tokens .

Despite being a lightweight model, it outperforms the larger Gemini 3 Flash on key agent benchmarks:

SWE-Bench Pro: 54.2% vs 49.6%

OSWorld-Verified: 74.0% vs 65.1%

The article suggests a "brain + workers" architecture : 3.6 Flash decomposes tasks, while multiple Flash-Lite instances execute them in parallel. A demo generated 25 production-ready web designs almost instantly. Flash-Lite is already live in the Gemini App and will roll into Google Search.

Gemini 3.5 Flash Cyber: Specialized Vulnerability Detection and Remediation

Flash Cyber is a dedicated cybersecurity model designed to find and fix vulnerabilities. Integrated into the CodeMender agent, multiple Cyber agents collaborate to achieve state-of-the-art on the CyberGym benchmark at lower cost than larger models.

The article notes an emerging asymmetry: AI discovers vulnerabilities faster than existing systems can patch them. Flash Cyber addresses this by automating remediation. However, the model is not publicly available at this time.

Gemini 4: "Most Aggressive Pre-training Ever" Underway

Alongside the three releases, Google confirmed it has started "the most aggressive pre-training in its history" targeting the next-generation Gemini 4 . Given Google's typical six-month training cadence, the article speculates Gemini 4 could appear by year-end.

Immediate Impact: Lower Costs for Agent Workloads

For practitioners paying for tokens and building agents, the key takeaway is cost reduction across three dimensions : lower per-token prices, fewer tokens consumed per task, and reduced overall agent execution cost. The flagship 3.5 Pro and the in-training Gemini 4 remain "futures"; the models available today deliver tangible savings.

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