Gemini 3.6 Flash - Google's Next-Generation AI Model

What is Gemini 3.6 Flash?

Gemini 3.6 Flash is Google DeepMind's latest flagship working model, positioned for large-scale AI agent scenarios. Compared to Gemini 3.5 Flash, it reduces output tokens by 17%, lowers the price to $7.5 per million tokens, and improves its DeepSWE benchmark from 37% to 49%, and its MLE-Bench benchmark from 49.7% to 63.9%. It also features comprehensive improvements in code, knowledge work, and computer operation capabilities, and includes built-in Computer Use tools, making it available to developers, enterprises, and general users.

Main functions of Gemini 3.6 Flash

• Main working modelPositioned as a workhorse, it is optimized for large-scale AI agent scenarios, reducing the number of output tokens by 17% compared to 3.5 Flash, resulting in a thinner bill.
• Code capability upgradeDeepSWE improved from 37% to 49%, and MLE-Bench improved from 49.7% to 63.9%, resulting in generated code that is closer to the production environment.
• Computer operating system built-inOSWorld-Verified rose from 78.4% to 83%, and "Computer Use" became an out-of-the-box tool for both APIs and enterprise editions.
• Knowledge work enhancementGDPval-AA v2 rose from 1349 to 1421, demonstrating outstanding performance in multimodal tasks such as document parsing, chart analysis, and report drafting.
• Security upgradeFrontier Safety offers enhanced protection, with a focus on preventing CBRN abuse and cyberattacks, and stronger resistance to jailbreaks.

Technical principles of Gemini 3.6 Flash

• Token efficiency optimizationBy improving the inference strategy, the model reduces unnecessary intermediate outputs when performing multi-step tasks. In DeepSWE scenarios, the output tokens can be reduced by 65%, allowing more effective work to be done at the same cost.
• Agent architecture streamlinedOptimize tool call chains, reduce inference steps and execution loop length, make code generation closer to production environment requirements, and reduce redundant modifications.
• Safety Alignment EnhancementDeploy an upgraded version of Frontier Safety protection, specifically aligned with two abuse scenarios: CBRN and cyberattacks, enhancing anti-jailbreak capabilities while reducing the false rejection rate for legitimate requests.

How to use Gemini 3.6 Flash

• Gemini AppTo use the Gemini app, open the app and select "3.6 Flash" from the model drop-down menu.
• Google AI StudioLog in to the studio and switch to Flash 3.6 in the model selector for development and testing.
• API callsSpecify the model name via the Gemini API gemini-3.6-flashInput price: $1.5/million tokens; Output price: $7.5/million tokens.
• Enterprise EditionWorkspace and Cloud enterprise users can enable it directly in the console; the Computer Use tool is ready to use out of the box.

The core advantages of Gemini 3.6 Flash

• More economicalOutput tokens are reduced by 17%, and in DeepSWE scenarios, even by 65%, resulting in lower single-task costs.
• FasterThe number of inference steps and tool calls is reduced, resulting in improved response speed.
• strongerThe Agent benchmarks, including code, machine learning engineering, computer operation, and knowledge work, have surpassed those of its predecessors.
• CheaperThe output price has dropped from $9 to $7.5 per million tokens, further improving its cost-effectiveness.
• SaferUpgraded Frontier Safety protection enhances jailbreak resistance and reduces false rejections.

Comparison of Gemini 3.6 Flash with similar competing products

Application scenarios of Gemini 3.6 Flash

• Software developmentDevelopers can leverage its enhanced coding capabilities to complete long-term software engineering tasks, automate code reviews, and generate test scripts.
• Machine learning researchResearchers can use it to assist in model training, experiment design, and data analysis in MLE-Bench type tasks.
• Intelligent document processingEnterprise users can perform multimodal analysis, chart extraction, and automatic report drafting on massive amounts of contracts, financial reports, and research reports.
• Computer AutomationIt automates browser operations, form filling, and system-level repetitive tasks through the built-in Computer Use tool.
• Enterprise Knowledge WorkIt can be embedded into toolchains such as Figma and JetBrains to assist in design iteration, code completion, and e-commerce data feature extraction.