Gemini 3 Flash Launches With Faster Performance and Smarter AI Capabilities
The Gemini 3 family is now complete with the launch of Gemini 3 Flash. Previously, the rollout began with the Pro model, and now the faster variant accelerates the transition. As a result, the new model arrives in the Gemini app and in search experiences. At the same time, developers gain immediate access through the Gemini API, Vertex AI, AI Studio, and Antigravity. Meanwhile, the larger generative models continue expanding in search, including both Gemini 3 Pro and its image component.
As expected, Gemini 3 Flash delivers higher speed and stronger capabilities than the earlier base model. According to published benchmarks, the model shows steady gains across academic and reasoning tests. For example, it outperforms the older 2.5 Flash in GPQA Diamond and MMMU Pro, and it even surpasses 3 Pro in one case. More notably, it posts a much larger improvement in Humanity’s Last Exam, which focuses on advanced domain knowledge. In that test, the score climbs to 33.7 percent without tool use, placing it just a few points behind the Pro model.
Stronger Coding, Accuracy, and Efficiency
In addition, Gemini 3 Flash demonstrates major progress in coding tasks. Over the past year, Pro models led code generation, yet Flash has rapidly closed the gap. On the SWE-Bench Verified test, it gains nearly 20 points over the 2.5 branch. Similarly, it answers general-knowledge questions more reliably. In the Simple QA Verified test, it reaches 68.7 percent, which sits only slightly below the Pro model. By contrast, the earlier Flash version scored just 28.1 percent.
Taken together, these results place Gemini 3 Flash much closer to the Pro tier than to the older family. However, it still emphasizes efficiency. The model runs workloads three times faster than Gemini 2.5 Pro, while also costing far less per token. One million input tokens cost $0.50, and one million output tokens cost $3. Although those prices rise from the previous Flash version, they remain well below Pro pricing.
Simplified Access Across Apps and Search
Meanwhile, the growing number of models had made the Gemini app harder to navigate. To address that issue, recent updates simplified and reorganized the settings. With this release, Gemini 3 Flash becomes the default model in both the app and the web interface. Consequently, the Fast and Thinking options now rely on this model. Gemini 3 Pro remains available under the Pro option, even though both versions can use the same simulated reasoning process.
In addition to app changes, Gemini 3 Flash rolls out immediately in search, primarily within AI Mode. Going forward, it serves as the default model there, which improves results for free users. AI Overviews will continue selecting models dynamically, although faster options will likely dominate. Still, more complex queries may rely on Pro models for subscribers. Finally, Gemini 3 Pro and its image model expand in AI Mode for US users, with higher usage limits reserved for paid plans.
