Google Gemini 1.5 AI model re-evolved: lower cost, better performance, and faster response

Google's Gemini 1.5 AI model has gotten a significant boost in evolutionary updates with the introduction of two brand new models: Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002. These new models not only enhance the quality and efficiency of the output, so that users can obtain more accurate and efficient services, but also significantly reduce the cost of use, making it more competitive in terms of economy. In addition, these updates provide users with higher rate limits, making the user experience smoother and faster when dealing with large amounts of data. Overall, the Gemini 1.5 update not only improves the performance of the AI model, but also creates greater value for users.

A brief introduction to the Gemini 1.5 AI model re-evolution

Functional re-evolution

Gemini-1.5-Pro-002

• You can seamlessly analyze, categorize, and summarize a large amount of content in a given prompt.
• You can perform highly complex understanding and reasoning tasks for different modes, including video.
• More relevant problem-solving tasks can be performed across longer blocks of code.
• High performance levels can be maintained with larger context Windows.

Gemini-1.5-Flash-002

• Visual information lookup: Use external knowledge combined with information extracted from input images or videos to answer questions.
• Object recognition: Answers questions related to the fine-grained recognition of objects in images and videos.
• Digital Content Understanding: Answer questions and extract information from visual content such as infographics, charts, graphs, tables and web pages.
• Structured content generation: Generate responses based on multimodal inputs in formats such as HTML and JSON.
• Captions and descriptions: Generate images and video descriptions with different levels of detail.
• Inference: The ability to combine and infer new information without memory or retrieval.

System configuration re-evolution

Filter settings reevolution

Building safe and reliable models has always been a priority. With the latest version of Gemini, the Gemini team has improved the model's ability to follow user instructions while maintaining security. And it will continue to provide a set of security filters for developers to apply to Google's model. For the models released today, these filters are not applied by default so that developers can determine the configuration that best suits their use cases.

Gemini 1.5 Flash-8B Experimental re-evolution

The team will release further improvements to the Gemini 1.5 model, released in August, named "Gemini-1.5-Flash-8B-EXP-0924." This improved version offers significant performance improvements in both text and multimodal use cases. It is now available through Google AI Studio and the Gemini API.

The benefits of the re-evolution of the Gemini 1.5 AI model

High speed limit

The Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002 AI models will also offer higher rate limits. The rate limit is the user's daily usage limit. With the 1.5 Flash model, users will get 2,000 requests per minute (RPMS), while the 1.5 Pro model will provide 1,000 RPMS.

High output speed, reduced delay

In addition to core improvements to the latest model, over the past few weeks we have reduced latency with 1.5 Flash and significantly increased output tokens per second, enabling new use cases with our most powerful model.

Stronger performance

In the more challenging MMLU-Pro benchmark test, the model's performance improved by about 7%. In the MATH and HiddenMath benchmarks, there was a significant 20% improvement in math performance. Visual and code-related tasks also improved, with 2-7% improvements in visual understanding and Python code generation assessments.

Lower cost

The most powerful 1.5 series model, Gemini 1.5 Pro, has 64% off input tokens, 52% off output tokens, and 64% off incremental cache tokens effective October 1, 2024. For tips with less than 128K tokens. Coupled with context caching, this will continue to reduce the cost of building with Gemini. Like the powerful XXAI with Claude3.5, GPT and dalle3 models, it is cheaper to use than other models, and the price is very competitive.

Other improvements

Google also upgraded its Gemini 1.5 experimental model, released in August, with an upgraded version of Gemini-1.5-Flash-8B-EXP-0924, with further enhancements for text and multimodal applications. The new Gemini model can be accessed through Google AI Studio, the Gemini API, and Vertex AI.

Conclusion

The Gemini 1.5 family of models is designed for general performance across a variety of text, code, and multimodal tasks. The continued advancement of the Gemini 1.5 model is intended to open up new possibilities for people, developers, and businesses to create, discover, and build with AI. Enable Gemini 1.5 to learn complex tasks faster and maintain quality, while improving training and service efficiency. In general, the overall quality of the model has improved, and the mathematics, long-term background and vision have made great progress.