Google's new AI model is cheaper and faster
What's the story
Google has launched its latest artificial intelligence (AI) model, Gemini 3 Flash, which will be the default option in the Gemini app and AI mode in search. The launch comes six months after the introduction of Gemini 2.5 Flash, which was also a major upgrade over its predecessor. While Gemini 3 Flash has a higher per-token cost than Gemini 2.5 Flash, it uses 30% fewer tokens on average for thinking tasks, potentially leading to overall cost savings for certain tasks.
Performance comparison
It outperforms predecessors and matches top models
The Gemini 3 Flash model has shown a huge improvement over its predecessor in benchmarks. It even matches the performance of other leading models such as Gemini 3 Pro and GPT-5.2 in some areas. For instance, it scored an impressive 33.7% on Humanity's Last Exam benchmark, which measures expertise across domains. This is significantly better than the score of its predecessor, Gemini 2.5 Flash (11%), and slightly below the score of the new entrant GPT-5.2 (34.5%).
Benchmark performance
Gemini 3 Flash excels in multimodality and reasoning
On the MMMU-Pro benchmark for multimodality and reasoning, the new model outperformed all competitors with a score of 81.2%. Google has made this model the default option in its global Gemini app, replacing the previous version. Users can still opt for the Pro model from a menu for math and coding queries.
Enhanced features
Capabilities and user experience
The new model is said to be good at identifying multimodal content and generating visual answers with elements like images and tables. It can also be used to create app prototypes in the Gemini app using prompts. Google has already made the Gemini 3 Pro available for search users in the US, along with broader access to its Nano Banana Pro image model.
Access expansion
It is now available for enterprises and developers
Google has made the Gemini 3 Flash model available to enterprises through Vertex AI and Gemini Enterprise. Companies like JetBrains, Figma, Cursor, Harvey, and Latitude are already using it. For developers, the company is offering a preview version via API and Antigravity, its new coding tool launched last month. The model excels at video analysis, data extraction, visual Q&A tasks due to its speed.