Google introduced Gemini 3.5 Flash and rolled out Gemini Spark as part of a broader update that aims to speed agentic workflows while keeping high quality, according to Google and company executives.
Google said Gemini 3.5 Flash matches frontier model intelligence on many fronts while delivering much lower latency, and it is four times faster on output tokens per second compared with other frontier models, as reported by Google.
According to Google, Gemini 3.5 Flash outperforms Gemini 3.1 Pro on complex coding and agent benchmarks, including Terminal-Bench 2.1 at 76.2 percent, GDPval-AA at 1656 Elo and MCP Atlas at 83.6 percent, and it leads multimodal understanding with 84.2 percent on CharXiv Reasoning, as reported by Google.
Google said the Flash model often completes long-horizon tasks at less than half the cost of other frontier models, and the company noted that 3.5 Pro is already in internal use and expected to be rolled out next month.
Developers can access 3.5 Flash through Google Antigravity, the Gemini API in Google AI Studio and Android Studio, while enterprises can use the Gemini Enterprise Agent Platform and Gemini Enterprise, Google said the model is now the default in the Gemini app and AI Mode in Search worldwide.
Google described Antigravity as a harness that lets 3.5 Flash run collaborative subagents to execute multi-step workflows, and the company cited examples such as automated renaming and categorization of assets, rapid game development, and legacy code migration.
Agent Rollouts Omni Features And Safety Measures
Gemini Spark is a cloud-based personal agent that runs continuously to take actions under user direction, and Google said it will start with trusted testers and then bring a beta to Google AI Ultra subscribers in the US, according to Google.
Google also introduced Gemini Omni, a multimodal world model intended to edit and generate video grounded in real-world knowledge, and the first variant Omni Flash is available today in the Gemini app, Google Flow and YouTube Shorts, Google said.
Google said Omni can accept images, audio, video and text to create or edit videos, and that each Omni-generated video will include a SynthID digital watermark to help users identify synthetic content.
The company said it strengthened cyber and CBRN safeguards for the Gemini 3.5 series and applied new safety training, mitigations and interpretability tools to reduce harmful outputs while avoiding mistaken refusals on safe queries, as reported by Google.
Google highlighted enterprise partners using 3.5 Flash and Antigravity, listing Shopify, Macquarie Bank, Salesforce, Ramp, Xero and Databricks as organizations piloting agentic workflows for tasks like forecasting, onboarding, OCR improvement and data diagnostics, according to Google.
