Google Unveils Gemini 3.5 Flash: A Leap Forward in AI Agents and Coding
Google has introduced Gemini 3.5 Flash, a faster and more affordable AI model for agents and coding, marking a significant advancement in intelligent agents.

Google Unveils Gemini 3.5 Flash: A Leap Forward in AI Agents and Coding">
['Google has made a significant announcement at its I/O conference in May 2026, introducing Gemini 3.5 Flash, the first model in the Gemini 3.5 series. This new model combines cutting-edge intelligence with action, enabling more sophisticated and autonomous AI agents. According to Google, Gemini 3.5 Flash represents a major leap forward for intelligent agents, offering enhanced performance and cost-effectiveness.', 'The Flash tier has historically been associated with faster and cheaper models, and Gemini 3.5 Flash is no exception.
It outperforms the previous premium tier, Gemini 3.1 Pro, on challenging benchmarks. Specifically, Gemini 3.5 Flash scores 76.2% on Terminal-Bench 2.1, which tests coding performance, and 1656 Elo on GDPval-AA, a measure of real-world agentic task performance. Additionally, it achieves 83.6% on MCP Atlas, which assesses scaled tool-use reliability, and 84.2% on CharXiv Reasoning, a benchmark for multimodal understanding.', 'In terms of performance and cost, Gemini 3.5 Flash is 4x faster on output tokens, with tasks often completed at less than half the cost.
The official pricing is $1.50 per million input tokens, while output tokens cost $9.00 per million. Cached input is priced at $0.15 per million. The model boasts a context window of 1,048,576 input tokens and a maximum output of 65,536 tokens, supporting text, image, audio, and video inputs.
The knowledge cutoff is January 2026, and dynamic thinking is enabled by default, allowing the model to auto-allocate more compute for harder problems.', 'Google has also introduced Managed Agents in the Gemini API, which enables the creation of full agents with a single API call. These agents can reason, use tools, and execute code within an isolated Linux container, persisting files and state across follow-up calls. This facilitates seamless multi-turn agent sessions.
Furthermore, Google has launched Antigravity, an agent-first development platform that streamlines the process of taking ideas to production-ready apps. The Antigravity 2.0 standalone desktop app allows for the orchestration of multiple agents running in parallel, with dynamic subagents handling parallelized workflows and scheduled tasks enabling background automation.', 'Several enterprise partners are already leveraging Gemini 3.5 Flash, including Shopify, which uses subagents in parallel for data analysis to power more accurate merchant growth forecasts globally. Macquarie Bank is piloting the model for customer onboarding, while Salesforce is integrating it into Agentforce to automate enterprise tasks using multiple subagents.
Source: MarkTechPost