Google's new Gemini Flash models: faster, cheaper, and one built for security
Google introduced Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, cutting token costs while adding a specialist model for its CodeMender security agent.
Three new Flash models
Google introduced three new Gemini models built on its Flash line: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber. Google frames all three around the same goal, saying developers and customers building production AI agents need "higher token efficiency, lower latency, and more reliable performance." The company says its Flash series is built to hit "the sweet spot of efficiency and quality" for scaling agentic workflows, building on the earlier Gemini 3.5 Flash. Beyond these three, Google says Gemini 3.5 Pro is still testing with partners and that it plans to make it broadly available as soon as it's ready, and that the team has already started what it calls its "most ambitious pre-training run yet," for Gemini 4.
3.6 Flash: more efficient, and stronger on benchmarks
Google calls 3.6 Flash its new workhorse model, meant to improve coding, knowledge work, and multimodal performance. On the Artificial Analysis Index, Google says 3.6 Flash cuts output token usage by 17% compared with 3.5 Flash, and in some benchmarks, such as DeepSWE by Datacurve, that gap reaches up to 65%, all while costing less per output token. It is also priced lower than 3.5 Flash, at $1.50 per million input tokens and $7.50 per million output tokens.
Beyond efficiency, Google reports separate accuracy gains for 3.6 Flash over 3.5 Flash on several benchmarks. On DeepSWE, it scored 49% versus 37% for 3.5 Flash, with fewer unwanted code edits and shorter execution loops, according to Google. On MLE Bench, a machine-learning research benchmark, it scored 63.9% versus 49.7%. On OSWorld-Verified, a computer-use benchmark, it scored 83.0% versus 78.4%, and Google says computer use is now a built-in client-side tool through the Gemini API and Gemini Enterprise. On GDPval-AA v2, a knowledge-work benchmark, it scored 1421 versus 1349. Google names Hebbia and Harvey as customers who have found 3.6 Flash especially capable at multimodal tasks such as document parsing, chart and data analysis, and report drafting.
3.5 Flash-Lite: built for speed and scale
Gemini 3.5 Flash-Lite is designed for low-latency tasks and for high-throughput workflows such as agentic search and document processing, according to Google. It is the fastest model in the 3.5 series, running at 350 output tokens per second as measured by Artificial Analysis, and is priced at $0.3 per million input tokens and $2.5 per million output tokens. Google says it significantly outperforms the earlier 3.1 Flash-Lite and calls it "a significant step up in coding and agentic tasks," citing scores of 54% versus 31% on Terminal-Bench 2.1, 72.2% versus 60.1% on GDM-MRCR v2 for long context, and 1140 versus 642 on GDPval-AA v2 for real-world task execution.
Google separately says that on several agentic and coding evaluations, 3.5 Flash-Lite even outperforms the larger 3 Flash model, scoring 54.2% versus 49.6% on SWE-Bench Pro and 74.0% versus 65.1% on OSWorld-Verified. Developers can also configure 3.5 Flash-Lite's thinking level, using minimal or low thinking for high-volume, low-cost tasks or higher thinking levels for multi-step subagent workloads, and it now supports computer use as a built-in tool as well.
3.5 Flash Cyber: a specialist for CodeMender
Gemini 3.5 Flash Cyber is a cyber-focused model built on top of 3.5 Flash and fine-tuned for finding and fixing security vulnerabilities at a lower price per token than larger models, Google says. It runs inside CodeMender, Google's code-security agent, where multiple 3.5 Flash Cyber agents work together to produce a single combined report. Google says the model reaches "competitive performance at the frontier" on the CyberGym benchmark, though it does not publish a numeric score there. Because of what Google calls the "dual-use nature" of this technology, access will be limited: 3.5 Flash Cyber will be exclusively available to governments and trusted partners through CodeMender, as part of a limited-access pilot program.
Safety guardrails on 3.6 Flash
Google says 3.6 Flash ships with enhanced Frontier Safety safeguards covering Chemical, Biological, Radiological, and Nuclear (CBRN) misuse and cyber-offense misuse. The company says these safeguards make the model "substantially more resistant to jailbreaks," while the model has also been trained to minimize refusals for beneficial uses.
Key facts
- Google introduced three new models: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber.
- 3.6 Flash cuts output token usage by 17% versus 3.5 Flash on the Artificial Analysis Index, up to 65% in benchmarks like DeepSWE, priced at $1.50 per million input tokens and $7.50 per million output tokens.
- 3.5 Flash-Lite is the fastest 3.5-series model, running at 350 output tokens per second, priced at $0.3 per million input tokens and $2.5 per million output tokens.
- 3.5 Flash Cyber runs inside CodeMender and will be exclusively available to governments and trusted partners through a limited-access pilot.
- 3.6 Flash ships with enhanced Frontier Safety safeguards for Chemical, Biological, Radiological, and Nuclear (CBRN) and cyber-offense misuse.
Check your understanding
By how much does 3.6 Flash cut output token usage compared with 3.5 Flash on the Artificial Analysis Index, and how high does that gap reach in benchmarks like DeepSWE? Which two customers does Google name as finding 3.6 Flash especially capable at multimodal tasks? Who will be able to access 3.5 Flash Cyber once it launches through CodeMender?