The company announced that it has released Gemini 3.5 Flash as the first in the Gemini 3.5 series, designed to support the execution of complex agent workflows. This model is now available to billions of users globally.
Gemini 3.5 Flash is described as providing frontier-level intelligence at the speed expected of the Flash series. Google DeepMind stated that the model outperforms Gemini 3.1 Pro in coding and agent-based benchmarks such as Terminal-Bench 2.1 and GDPval-AA. Additionally, it reported a multimodal understanding score of 84.2% on CharXiv Reasoning.
In terms of processing speed, the model reportedly recorded output token counts (per second) four times higher than other frontier models. It is designed to resolve the tradeoff between quality and latency, enabling support for long-term agent tasks. The company claims that tasks that previously took developers or auditors days to weeks to complete can now be finished in a short time and at a low cost.
When combined with the Antigravity harness, it functions as a deployment engine for collaborative sub-agents. It is said to reliably execute multi-step workflows and coding tasks under supervision while maintaining frontier performance. Specifically, the company demonstrated the automatic classification of unstructured assets and game development based on the AlphaZero paper.
Enhancements to the multimodal foundation have enabled the generation of richer, more interactive Web UIs and graphics. On AI Studio, the model has realized the creation of animations for research papers and the generation of hardware interactions from text descriptions.
In practical applications, Shopify is executing sub-agents in parallel to analyze complex data and improve the accuracy of growth forecasts for merchants. It has been reported that Macquarie Bank has begun pilot operations aimed at accelerating customer onboarding.
Source: Gemini 3.5 Flash: frontier intelligence with action (HN 180pt, 1 comment) (HN Search (backfill), 2026-05-20)