Earlier today, we saw the launch of DeepSeek's new large language model, DeepSeek-V3-0324. Now, Google has announced Gemini 2.5, a collection of thinking models that it claims are its most intelligent. As Google puts it:
Gemini 2.5 is a thinking model, designed to tackle increasingly complex problems. Our first 2.5 model, Gemini 2.5 Pro Experimental, leads common benchmarks by meaningful margins and showcases strong reasoning and code capabilities.
Gemini 2.5 Pro Experimental is available in Google AI Studio as well as the Gemini app as long as you're subscribed to Gemini Advanced. There are also plans to bring it to Vertex AI, though Google did not reveal a time frame.
From the benchmarks Google showed, Gemini 2.5 Pro Experimental leads in math and code generation benchmarks, like AIME 2025 and LiveCodeBench v5. The company says that 2.5 represents a 'big leap' in coding performance over 2.0, making it better at creating web apps, editing code, and more.
Here's a table showing some information about Gemini 2.5 Pro:
| Model deployment status | Experimental |
|---|---|
| Supported data types for input | Text, Image, Video, Audio |
| Supported data types for output | Text |
| Supported number tokens for input | 1M |
| Supported number tokens for output | 64K |
| Knowledge cutoff | January 2025 |
| Tool use |
|
| Best for |
|
| Availability |
|
For now, there's no pricing, but expect more information regarding that in the coming weeks.
Over the last few months, Google added a couple of interesting Gemini-based models, like Gemini Robotics, a Gemini 2.0 model for robots. The consumer-facing Gemini App recently got Deep Research, a feature that uses AI to conduct research for you.
1 Comment - Add comment