Google has introduced Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber, three new AI models designed for AI agent workflows with improvements in token efficiency, latency and performance.
According to Tulsee Doshi, Senior Director of Product Management at Google, on behalf of the Gemini team, developers and customers building production AI agents require higher token efficiency, lower latency and reliable performance. Built on Gemini 3.5 Flash, the new models target coding, knowledge work, multimodal processing, agent workflows and cybersecurity applications.
Gemini 3.6 Flash
Gemini 3.6 Flash is designed for coding, knowledge work and multimodal tasks. According to the Artificial Analysis Index, the model reduces output token usage by 17% compared to Gemini 3.5 Flash. Google also said DeepSWE by Datacurve observed improvements of up to 65% in some benchmarks. The model uses fewer reasoning steps and tool calls for multi-step workflows.
Pricing:
- $1.50 per 1 million input tokens
- $7.50 per 1 million output tokens
Compared to Gemini 3.5 Flash, Google reported improvements across coding, computer use and knowledge work.
| Benchmark | Gemini 3.6 Flash | Gemini 3.5 Flash |
|---|---|---|
| DeepSWE | 49% | 37% |
| MLE Bench | 63.9% | 49.7% |
| OSWorld-Verified | 83.0% | 78.4% |
| GDPval-AA v2 | 1421 | 1349 |
Google said DeepSWE also showed higher precision with fewer unwanted code edits and reduced execution loops. Computer use is available as a built-in client-side tool through the Gemini API and Gemini Enterprise.
The company also said customers including Hebbia and Harvey have used Gemini 3.6 Flash for multimodal tasks such as document parsing, chart and data analysis, and report drafting.
Safety features
Gemini 3.6 Flash includes enhanced Frontier Safety safeguards covering:
- Chemical
- Biological
- Radiological
- Nuclear (CBRN)
- Cyber offense misuse
Google said the safeguards improve resistance against jailbreak attempts while the model has been trained to minimize refusals for beneficial use cases.
Gemini 3.5 Flash-Lite
Gemini 3.5 Flash-Lite is designed for low-latency and high-throughput workloads, including agentic search and document processing. According to Artificial Analysis, the model delivers 350 output tokens per second.
Pricing:
- $0.30 per 1 million input tokens
- $2.50 per 1 million output tokens
Compared to Gemini 3.1 Flash-Lite, Google said the model improves coding, long-context processing and real-world task execution.
| Benchmark | Gemini 3.5 Flash-Lite | Comparison |
|---|---|---|
| Terminal-Bench 2.1 | 54% | 31% |
| GDM-MRCR v2 | 72.2% | 60.1% |
| GDPval-AA v2 | 1140 | 642 |
| SWE-Bench Pro | 54.2% | 49.6% (Gemini 3 Flash) |
| OSWorld-Verified | 74.0% | 65.1% (Gemini 3 Flash) |
Google said developers can configure the model with minimal or low thinking levels for lower-latency, lower-cost workloads, or use higher thinking levels for multi-step subagent workflows. Gemini 3.5 Flash-Lite also includes computer use as a built-in tool for AI agent tasks across supported surfaces.
Gemini 3.5 Flash Cyber with CodeMender
Google also introduced Gemini 3.5 Flash Cyber, a cybersecurity-focused model integrated with the CodeMender code security agent.
Built on Gemini 3.5 Flash, the model is fine-tuned to identify, validate and fix software vulnerabilities. Within CodeMender, multiple Gemini 3.5 Flash Cyber agents work together to generate a combined security report, with Google reporting competitive performance on the CyberGym benchmark.
Due to the dual-use nature of cybersecurity AI, Gemini 3.5 Flash Cyber will initially be available only to governments and trusted partners through CodeMender as part of a limited-access pilot program.
Availability
Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are available starting today through:
- Gemini API via Google AI Studio and Android Studio
- Gemini Enterprise Agent Platform
Gemini 3.6 Flash is also available in:
- Google Antigravity
- Gemini Enterprise app
Gemini 3.5 Flash-Lite is rolling out through:
- Gemini app
- Google Search
Google also confirmed that Gemini 3.5 Pro is currently being tested with partners before broader availability. The company has also started its next-generation pre-training run for Gemini 4.