Google introduced the Gemini 3.5 model family at Google I/O 2026, with Gemini 3.5 Flash becoming generally available first. The company described it as its strongest model for agentic tasks and coding, built to handle demanding workflows while keeping latency and cost under control. Google said the model delivers near-flagship reasoning performance and, in complex agentic workflows, code generation, and multimodal tasks, can outperform the earlier Gemini 3.1 Pro.
A faster Flash model aimed at agentic workflows and coding
The Flash line had previously been positioned as a lightweight and fast foundation model, but Google is now pushing Gemini 3.5 Flash much higher in the stack. According to the company, the model combines stronger reasoning with the speed expected from the Flash series. Google said processing speed can reach 4x that of peer frontier models, while compute cost is often less than half of competing offerings. That makes it relevant for applications that rely on frequent calls, long context windows, and coordinated multi-agent execution.
Demo used 93 sub-agents to build a runnable operating system
The most eye-catching part of the presentation was a live demonstration focused on complex software engineering. Google said its team used Gemini 3.5 Flash together with the Antigravity 2.0 framework, deploying 93 parallel sub-agents. Over the course of 12 hours, those agents processed 2.6 billion tokens and built a runnable operating system from scratch. The total API bill for the demo, according to Google, came in at under $1,000.
That showcase was meant to make a specific point. Google is presenting Gemini 3.5 Flash as more than a low-cost, high-speed model; it is positioning the release for heavier engineering tasks that involve orchestration, decomposition, and repeated execution across many agents. For teams building software with large token throughput and sustained API usage, those figures are likely to be the headline metrics.
Pricing details released as Google maps out the 3.5 lineup
Google also disclosed API pricing for Gemini 3.5 Flash. Input pricing is listed at about $1.50 per million tokens, while output pricing is about $9.00 per million tokens. The company noted that pricing is slightly higher than the previous Gemini 3 Flash, but framed the new version as a stronger value proposition because of its reasoning and agent coordination capabilities.
Google added that the Gemini 3.5 rollout is only beginning. Alongside the Flash release, the company plans to introduce Gemini 3.5 Pro for stronger reasoning performance and a lighter Lite variant, giving developers more options across different use cases.