Google has announced significant updates of its Gemini models, aimed at making advanced AI opportunities more accessible and profitable for programmers around the world. Two new models ready for production: Gemini-1.5-PRO-002 and Gemini-1.5-Flash-002 contain a significant improvement in speed and performance.
Key updates:
- Price reduction: 64% Lowering the prices of input tokens, 52% of production tokens and 64% in incremental buffered tokens for Gemini 1.5 Pro.
- Increased speed limits: The 1.5 Flash speed limit has doubled to 2000 rpm, and for 1.5 Pro it will almost get up to 1000 rpm.
- Speed improvement: Gemini models now offer 2x faster output and 3x lower delay, which makes it easier for programmers to implement high -performance artificial intelligence in real time.
The Gemini 1.5 series has been designed for a wide range of tasks, including text, code and multimodal applications. These models can support large inputs, such as 1000-page PDF files and hour movies, offering increased performance in key areas:
- 7% improvement in relation to MMLU-PRO, which assesses the understanding of AI.
- 20% improvement of complex mathematical tasks, such as mathematics and hidden.
- Better results for visual understanding and production of Python code.
Responding to the feedback of programmers, Gemini 1.5 models now generate a more concise result-5-20% shorter than previous versions. This is especially useful for summary and extraction of information, reducing the total costs while maintaining transparency and accuracy.
The new models are equipped with updated safety filters, enabling programmers to adapt them on the basis of specific needs. The default filters have been corrected to balance compliance with the user manual and security.
The improved experimental version, Gemini-1.5-Flash-8B-EXP-0924 was also released. This model includes significant updates in both text and multimodal text and is available via Google AI Studio and Gemini API.
The latest improvements make Gemini 1.5 models faster, more profitable and better suitable for a wide range of applications. Developers can access these models for free via Google to learnwhile larger organizations and customers of Google Cloud can use them Vertex AI.
For more information, visit the Google blog for programmers.
