Google: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite reach general availability with 17% fewer tokens and a new Cyber model

What is Gemini 3.6 Flash and how does it differ from its predecessor?

Gemini 3.6 Flash is Google’s fast, lower-cost model within the Gemini family — the Flash tier targets tasks that need lower latency and cost than the “Pro” models, with only a small quality trade-off. Google has declared the model General Availability (GA), meaning it has left the beta phase and is available to all developers for production use via the API. Pricing is $1.50 per million input tokens and $7.50 per million output tokens. The key change versus Gemini 3.5 Flash is efficiency: the new model uses 17% fewer output tokens for comparable tasks, directly lowering the cost per query. For companies building production AI products, General Availability status also means contractual guarantees of API stability — no sudden changes in model behavior of the kind possible during the preview phase.

Benchmarks: big jumps on DeepSWE, MLE-Bench and OSWorld

On the DeepSWE benchmark, which measures the ability to solve software tasks autonomously, Gemini 3.6 Flash scores 49% versus 37% for Gemini 3.5 Flash. On MLE-Bench, a benchmark for machine learning tasks, the score rises from 49.7% to 63.9%. On OSWorld-Verified, a test of computer interface control, the model scores 83.0% versus the earlier 78.4%. All three jumps show a consistent improvement in agentic capability — not just response speed, but also accuracy in multi-step tasks such as writing code and operating an interface. Together, DeepSWE, MLE-Bench and OSWorld-Verified cover three distinct agentic tasks — writing code, machine-learning engineering, and navigating a computer’s graphical interface — so a consistent score increase across all three suggests the improvement is not narrowly specialized to one task type, but stems from generally better model reasoning.

Gemini 3.5 Flash-Lite: the cheapest tier with a big jump on Terminal-Bench

Alongside the larger Flash model, Gemini 3.5 Flash-Lite — the cheapest and fastest tier in the Gemini family — also reaches general availability: $0.30 per million input tokens, $2.50 per million output tokens, at a speed of 350 tokens per second. On Terminal-Bench 2.1, a test of executing terminal commands, Flash-Lite scores 54% versus 31% for its predecessor — nearly double the improvement. On SWE-Bench Pro, the score rises from 49.6% to 54.2%. For developers choosing between the two tiers, the trade-off is clear: Flash-Lite sacrifices some raw capability for lower cost and faster generation, while the full Gemini 3.6 Flash retains greater precision at a higher cost per token.

Cyber model for governments only, and the announced Gemini 3.5 Pro

Within the CodeMender tool, Google also announced Gemini 3.5 Flash Cyber, a specialized model for automatically fixing security vulnerabilities in code, which will soon be available through a limited-access pilot program for governments and partner organizations, not the wider developer community. Google simultaneously announced Gemini 3.5 Pro, currently in the partner-testing phase ahead of a broader launch. That means the Gemini family gets two General Availability announcements in a single day (3.6 Flash and 3.5 Flash-Lite), one announced cybersecurity model, and one Pro model in preparation — a move that spans the range from the cheapest tier to a specialized, high-security tier.