Google Gemini 3.5 Flash: Major Update for AI Agents and Programming Efficiency

Technical performance and speed

Gemini 3.5 Flash is designed to provide intelligence comparable to larger flagship models while maintaining the low latency typical of the Flash series. According to performance data, the model is four times faster than other Frontier models in terms of the number of tokens issued per second.

In standardized tests, the model shows significant improvements in programming and agent capabilities. It achieved a score of 76.2% on Terminal-Bench 2.1 and 83.6% on MCP Atlas, while on GDPval-AA it achieved 1656 Elo. Furthermore, its multimodal reasoning skills are underlined by a score of 84.2% on CharXiv Reasoning, giving it a good position in the Artificial Analysis Index in terms of both quality and speed.

Scaling agentic workflows

The model was developed specifically for tasks with a long time horizon that traditionally require a high level of manual effort. By automating the planning, creation and iteration phases of a project, Gemini 3.5 Flash can reduce the time required for audits or software development from weeks to hours. These operational improvements are accompanied by a reduction in costs, as tasks are often completed at less than half the cost of other top models. Using the Antigravity Harness, Gemini 3.5 Flash can manage collaborative subagents. This allows the system to perform multi-step workflows under human supervision, such as automatically categorizing and renaming unstructured digital assets.

Integration in companies and for consumers

Industrial use can already be observed in several industries. Financial institutions and fintech companies are using the model to automate complex workflows, while data science teams are using it to gain insights from complex data environments. Shopify, in particular, uses parallel subagents to analyze long-term data for global merchant growth forecasts.

For general users, Gemini 3.5 Flash is now the default model for the Gemini app and AI mode in Google Search. This update enables the introduction of Gemini Spark, a personal AI agent that can manage digital tasks around the clock. Gemini Spark is currently being rolled out to trusted testers, with a beta version planned for Google AI Ultra subscribers in the United States next week.

In addition, the model enhances the search experience by introducing information agents and dynamic generative UI elements, such as interactive visual explanations for complex scientific patterns.

Security and availability

Gemini 3.5 was developed in accordance with the Frontier Safety Framework. This included the implementation of enhanced safeguards against cyber threats and CBRN risks. To reduce the likelihood of malicious content generation or false rejection of search queries, Google has integrated advanced security training and interpretability tools that allow the model's internal reasoning to be analyzed before a response is issued.

Gemini 3.5 Flash is accessible via the Gemini API in Google AI Studio and Android Studio, Google Antigravity and the Gemini Enterprise Agent Platform. It is also available to all users via the Gemini app and Google Search.