Google has unveiled Gemini 3 Flash, a state-of-the-art AI model designed to deliver frontier-level intelligence coupled with high speed and low cost. This latest iteration in the Gemini lineup achieves impressive scores, including 90.4% on the GPQA Diamond benchmark and 81.2% on the MMMU Pro benchmark. In addition, it accomplishes this while being three times faster than its predecessor, Gemini 2.5 Pro, and utilizing 30% fewer tokens.
One of the standout features of Gemini 3 Flash is its enhanced capabilities in coding tasks, as it scores 78% on the SWE-bench Verified. Additionally, the model showcases robust multimodal functionalities that include video analysis and visual Q&amp;A. Notably, advanced visual reasoning is powered by the agentic vision feature, which operates in a think-act-observe loop.
Agentic vision enables the model to perform multi-step analysis of images, generating precise Python code for image manipulation tasks such as cropping, zooming, and annotating. Before providing final outputs, this approach allows the model to verify visual details, enhancing accuracy on vision benchmarks by 5-10%. This detailed inspection through deterministic code effectively handles complex visual arithmetic tasks.
Gemini 3 Flash is available through multiple platforms including the Gemini API, Google AI Studio, and Vertex AI, and it’s being introduced as the default model across the Gemini app and Google’s AI Mode in Search worldwide. The pricing is set at an accessible $0.50 for 1 million input tokens and $3 for 1 million output tokens. Future plans entail extended support across the entire Gemini family, solidifying Gemini 3 Flash as a cornerstone of Google’s AI developments.

Collections

Gemini 3 Flash is Google's newest AI model offering frontier-level intelligence with 3x faster performance and 30% fewer tokens than Gemini 2.5 Pro. It achieves 90.4% on GPQA Diamond and 78% on SWE-bench Verified benchmarks. The standout feature is agentic vision, which uses a think-act-observe loop to perform multi-step image analysis and generate Python code for image manipulation, improving vision benchmark accuracy by 5-10%. Available through Gemini API, Google AI Studio, and Vertex AI at $0.50 per million input tokens.

Introducing Google’s Gemini 3 Flash with Advanced Agentic Vision