Gemini 3.5 Flash might be fast enough for gen AI to make sense

Google has unveiled Gemini 3.5 Flash, its latest AI model, which promises significant advancements in both efficiency and intelligence. This release marks a rapid evolution, coming shortly after the 3.0 and 3.1 families, and is now being integrated across a wide range of Google products. The company claims this new model surpasses its previous-generation Pro model.

Gemini 3.5 Flash stands out for its ability to deliver frontier-level intelligence while operating with remarkable efficiency. This makes it particularly well-suited for complex agentic tasks, which have historically been resource-intensive for generative AI. The model can output nearly 300 tokens per second, a notable improvement over larger frontier models that operate at a quarter of that speed.

One of the key benefits of this increased efficiency is cost savings. Google estimates that companies heavily reliant on AI tokens could save up to a billion dollars annually by migrating to Gemini 3.5 Flash. This addresses a major challenge in the generative AI space, where the high operational costs of complex agentic experiences have been a barrier to widespread adoption.

The model demonstrates impressive performance in code generation and handling general tasks in real computing environments. Benchmarks like Terminal Bench, SWE-Bench Pro, and OSWorld-Verified show substantial improvements over older Flash models and even slight gains against Gemini 3.1 Pro, placing it in a similar performance bracket as OpenAI's more expensive GPT 5.5.

Internally, Google has already deployed Gemini 3.5 Flash, with employees reporting a "massive jump" in coding performance. The Antigravity IDE is being updated to version 2.0 with support for the new model, enabling parallel workflows and sub-agents. Gemini 3.5 Flash will also become available across various Google platforms, including the Gemini app, API, AI Studio, Android Studio, and all enterprise products.

Beyond the model itself, Google is introducing Gemini Spark, a dedicated AI agent running 24/7 in the cloud. Spark leverages Gemini 3.5 Flash to manage multiple agentic workflows across a user's entire Google footprint, including Drive and Gmail. This allows it to perform tasks like summarizing meetings, creating email digests, and proactively seeking user approval for high-stakes actions, effectively acting as a personal digital assistant.