Google Launches Gemini 3.6 Flash, Flash Lite and Cyber Models

Abstract digital neural network visualization with interconnected nodes and glowing lines representing artificial intelligence

In brief

  • Google launches Gemini 3.6 Flash, Flash Lite, and Cyber models for AI agents
  • Gemini 3.6 Flash reduces output tokens by 17% at $1.50 per million input tokens
  • Flash Lite generates 350 output tokens per second at $0.30 per million input tokens
  • Flash Cyber specializes in software vulnerability detection via CodeMender pilot

Gemini 3.6 Flash: Coding and multimodal gains

Gemini 3.6 Flash serves as the main model in the release, offering stronger coding, knowledge work and multimodal performance while consuming fewer tokens than Gemini 3.5 Flash. The model uses 17% fewer output tokens on the Artificial Analysis Index and reached token reductions as much as 65% in certain DeepSWE tests.

Performance gains are substantial. On coding benchmarks, Gemini 3.6 Flash scored 49% on DeepSWE, compared with 37% for Gemini 3.5 Flash. Its MLE Bench score increased to 63.9% from 49.7%. For computer use tasks, Gemini 3.6 Flash scored 83% on OSWorld Verified, compared with 78.4% for the previous model.

The model supports text, images, audio and video, with a context window of up to one million tokens. Pricing is competitive: $1.50 per million input tokens and $7.50 per million output tokens. The model is particularly suited for agentic workflows, coding tasks and long running enterprise processes.

Flash Lite: Speed and efficiency

Flash Lite targets developers who prioritize cost and latency. The model generates around 350 output tokens per second, making it among the fastest available. It is priced at $0.30 per million input tokens and $2.50 per million output tokens.

Benchmark improvements are significant. Gemini 3.5 Flash Lite scored 54% on Terminal Bench 2.1, compared with 31% for Gemini 3.1 Flash Lite. On other metrics, it reached 72.2% on GDM MRCR v2, up from 60.1% and scored 1,140 on GDPval AA v2, compared with 642. Flash Lite also outperformed the larger Gemini 3 Flash on several coding and agent benchmarks, including SWE Bench Pro and OSWorld Verified.

Flash Cyber: Cybersecurity focus

Gemini 3.5 Flash Cyber is specialized for finding, validating and fixing software vulnerabilities. The model operates through CodeMender, Google's code security agent. Flash Cyber will initially be available only to governments and trusted partners through a limited CodeMender pilot.

Availability and roadmap

Gemini 3.6 Flash and Gemini 3.5 Flash Lite are available through the Gemini API and Google AI Studio. Google's roadmap includes ongoing development of larger models. Google confirmed that Gemini 3.5 Pro is being tested with partners and it has begun its most ambitious pretraining run yet for Gemini 4.