Google roles out Gemini 1.5 upgrades
Upgrades to Google's flagship LLM!
Created on September 25|Last edited on September 25
Comment
Google has introduced two updated production-ready models: Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002. These updates focus on improving performance while significantly reducing costs. Notably, the pricing for the 1.5 Pro model has been reduced by more than 50% for prompts under 128K tokens, and rate limits have doubled for both 1.5 Flash and 1.5 Pro models. These developments make the models faster and more affordable for developers working with large-scale, AI-driven tasks.
Performance Improvements in Text, Code, and Vision
The latest Gemini 1.5 models are designed for a wide array of applications, including text, code, and multimodal tasks. The models have demonstrated improved performance in benchmarks, such as a 20% improvement on MATH and HiddenMath evaluations, which test the models' ability to handle complex mathematical problems. The models also show gains in Python code generation and visual understanding tasks, making them more efficient across diverse use cases like synthesizing long PDFs or analyzing video content.
Increased Efficiency and Reduced Response Lengths
Both models now generate more concise outputs, addressing feedback from developers who requested more efficient responses. For tasks like summarization and question answering, the models’ outputs are now 5-20% shorter than previous iterations. This reduces both costs and processing times, making the models more practical for production-level use.
Price Reductions and Rate Limit Increases for Gemini 1.5 Pro
One of the most significant updates is a 64% price reduction for input tokens and a 52% cut for output tokens in the Gemini 1.5 Pro model. Additionally, rate limits for paid tiers have been doubled, allowing developers to process more requests with lower latency. These improvements make the Gemini models a cost-effective solution for businesses requiring high-performance AI capabilities.
Gemini 1.5 Flash-8B Experimental Model
Google is also rolling out an improved version of the Gemini 1.5 Flash model, known as Flash-8B-Exp-0924. This experimental model provides enhanced performance for both text and multimodal tasks and is available for developers to explore through Google AI Studio and the Gemini API.
These updates position the Gemini 1.5 models as an increasingly powerful tool for developers, with enhanced performance across a wide range of tasks while becoming more cost-effective and accessible.
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.