Gemini 3 Flash vs. GPT-5.2: The New Benchmark Battle
Also Read | Flipkart End of Season Sale: Apple iPhone 16 Pro Sees Massive Price Slash
Google just took a massive swing at OpenAI today. The thing is, they’ve officially launched Gemini 3 Flash, and it’s already the default model for everyone using the Gemini app. Actually, it’s a direct response to the recently released GPT-5.2. Specifically, it’s designed to be the “workhorse” of the ecosystem. As a result, if you’re using Gemini today, your answers are now coming from a model that is three times faster than the previous generation. Consequently, Google is trying to steal back the “frontier” crown with raw speed (let’s be real, the competition is getting wild).
And here’s the kicker. This smaller model is actually outperforming “Pro” models on some of the hardest tests. In fact, on the MMMU-Pro benchmark, it hit a score of 81.2%, edging out both Gemini 3 Pro and GPT-5.2. Basically, it’s not just fast; it’s genuinely smart. Instead of struggling with complex PhD-level questions, it scored 33.7% on Humanity’s Last Exam. Furthermore, it handles multimodal tasks like a pro. You can upload a video of your golf swing or a messy hand-drawn sketch, and it just gets it.
Also Read | Flipkart End of Season Sale: Apple iPhone 16 Pro Sees Massive Price Slash
[Field Note: I checked the SWE-bench coding scores. Flash hit 78%, which actually beats out Gemini 3 Pro’s 76.2%. It’s a coding beast (those too).]
-
Global Rollout: It’s live now in the Gemini app and Search AI mode.
-
Multimodal Powers: Handles audio, video, and images simultaneously for real-time analysis.
-
App Prototyping: You can now use a single prompt to generate a functioning app prototype directly in the chat.
Moreover, the pricing for developers is a bit of a shift. Specifically, it costs $0.50 per 1 million input tokens. Actually, that is slightly more expensive than Gemini 2.5 Flash. As a result, Google claims you’ll still save money. In fact, because the model is more efficient, it uses 30% fewer tokens for complex “thinking” tasks. Consequently, you get more bang for your buck on bulk workflows. Basically, it’s built for companies like Figma and JetBrains that need to process trillions of tokens every single day.
The thing is, OpenAI is reportedly in “Code Red” mode. Actually, Sam Altman sent a memo to his team after seeing ChatGPT traffic dip while Google’s market share grew. Specifically, the battle for user preference is heating up. Instead of slowing down, Google is pushing Gemini 3 Pro to everyone in the U.S. for search today as well. And then Y followed. The Nano Banana Pro image model is also rolling out to more users. Consequently, the AI war of 2025 is officially in overdrive….
Also Read | Flipkart End of Season Sale: Apple iPhone 16 Pro Sees Massive Price Slash
