What you need to know
- Google is now rolling out Gemini 3.5 Flash globally across the Gemini app, AI Mode in Search, Google AI Studio, Android Studio, and enterprise platforms.
- Google says Gemini 3.5 Flash beats Gemini 3.1 Pro in several coding and reasoning benchmarks, including Terminal-Bench 2.1, GDPval-AA, and MCP Atlas.
- The company claims Gemini 3.5 Flash generates responses four times faster than rival frontier AI systems while maintaining high-end performance.
Google is pushing Gemini into a much more agentic future, and this time around the company wants to sell more than just raw speed. Google says its latest AI model, called Gemini 3.5 Flash, is designed to address complex workflows, multi-step tasks, and coding jobs that previously needed much larger and slower models.
Unveiled at today's Google I/O event, Gemini 3.5 Flash is now rolling out globally through the Gemini app and AI Mode in Google Search, with developers and enterprise customers able to access through Google Antigravity, Google AI Studio, Android Studio, and Gemini Enterprise tools.
3.5 Pro is coming next month
Google has also teased Gemini 3.5 Pro, but that version is still in testing internally and is expected to launch sometime next month.
To give you some context, Google's Flash models have always focused on performance versus speed and not on chasing pure benchmark dominance. Gemini 3.5 Flash, though, sounds like Google's attempt to eliminate that compromise altogether. The company claims that the model does better than Gemini 3.1 Pro on various tests for decision-making and coding, like Terminal-Bench 2.1, GDPval-AA, and MCP Atlas, and it also does well on tests for understanding different types of information, such as CharXiv Reasoning.
Google also claims that the model produces output four times faster than competing frontier AI systems. The company says Gemini 3.5 Flash is particularly strong at “long-horizon” tasks, meaning workflows that require sustained reasoning across multiple steps. In practical terms, Google believes the model can help developers build apps faster, maintain codebases more efficiently, and even assist financial teams with document preparation in a fraction of the time older workflows required.
A key part of that strategy is Google Antigravity, the company's agentic development platform. When paired with Gemini 3.5 Flash, Google says the platform can deploy collaborative subagents capable of tackling large-scale workloads under human supervision. The company also says the enhanced model’s stronger multimodal capabilities allow it to generate richer web interfaces and more dynamic graphics.
Google is already building those agentic ambitions right into consumer products. At Google I/O, the company announced Gemini Spark, a personal AI agent built on top of the new model, as Gemini 3.5 Flash rapidly becomes the model of choice for powering the Gemini app and AI Mode in Search globally.
According to Google, Gemini Spark is designed to run continuously in the background, helping users manage digital tasks and take actions under their direction. Starting today, trusted testers are getting access, and Google AI Ultra subscribers in the US should see the beta rollout next week.
As AI systems gain more autonomy, Google will have to face more scrutiny on safety. The company says Gemini 3.5 was developed under its Frontier Safety Framework and has enhanced cyber and CBRN protections. As a result, the model is better at refusing to produce harmful responses while reducing unnecessary refusals to safe prompts. Some of that is reportedly due to newer interpretability tools that help explore the model’s internal reasoning before it answers.
Gemini 3.5 Flash is accessible across Google’s AI ecosystem, including the Gemini app, AI Mode in Search, Google AI Studio, Android Studio, and enterprise platforms.
Android Central's Take
Tech companies love to promise frictionless automation until the moment the AI breaks something important or misunderstands a simple request. The Gemini 3.5 Flash sounds impressive on paper, particularly in terms of its speed and coding improvements, but putting more control in the hands of AI systems also means users will have to trust Google a lot more than they used to. And honestly, after years of companies trying to hype up AI features that sometimes feel half-baked at launch, I think a little skepticism is still healthy here.
