The company commenced the rollout with the immediate global release of Gemini 3.5 Flash, a high-speed model optimized for coding and long-horizon tasks. A larger, more powerful model, Gemini 3.5 Pro, is currently being utilized internally and is scheduled to launch next month.
According to technical specifications released by Google, Gemini 3.5 Flash delivers output token speeds four times faster than competing frontier models. On specialized coding and agentic benchmarks, the model outperformed Google's previous Gemini 3.1 Pro, securing scores of 76.2% on Terminal-Bench 2.1, 1,656 Elo on GDPval-AA, and 83.6% on MCP Atlas. It also registered an 84.2% score on the CharXiv Reasoning benchmark for multimodal understanding.
Gemini 3.5 Flash has been integrated directly into Google's consumer ecosystem, becoming the default model for the Gemini app and the AI Mode in Google Search. For developer and corporate deployment, the model is available via Google Antigravity, the company's agent-first development platform, as well as the Gemini API in Google AI Studio, Android Studio, and Gemini Enterprise.
Google showcased several commercial integrations already piloting the architecture. Salesforce is integrating the model into Agentforce to manage multi-turn tool calling, while Shopify is running parallel subagents to forecast global merchant growth. Other enterprise partners using the technology to automate lengthy administrative operations include Macquarie Bank, Ramp, Xero, and Databricks.
Additionally, Google announced a limited rollout of "Gemini Spark," a personal AI assistant powered by Gemini 3.5 Flash that operates continuously to complete tasks under user direction. The agent is available to trusted testers today, with a beta release scheduled for Google AI Ultra subscribers in the United States next week.
Addressing safety and compliance, Google stated that Gemini 3.5 was developed in accordance with its Frontier Safety Framework. The system features enhanced cybersecurity and chemical, biological, radiological, and nuclear (CBRN) safeguards. The training phase also incorporated advanced interpretability tools to audit the model's internal reasoning before outputs are generated, reducing both harmful content generation and mistaken refusals.
I truly appreciate you spending your valuable time here. To help make this blog the best it can be, I would love your feedback on this post. Let me know in the comments: How could this article be better? Was it clear? Did it have the right amount of detail? Did you notice any errors?
If you found any of the articles helpful, please consider sharing it.