Google has unveiled two updated production-ready models in its Gemini series: Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002. These models come equipped with numerous enhancements designed to improve performance, reduce costs, and expand usability for developers.
Today, we’re excited to release two new, production-ready versions of Gemini 1.5 Pro and Flash. 🚢
According to Google’s announcement, the new Gemini models boast a ~7% increase in MMLU-Pro scores and a ~20% improvement in math-related benchmarks, specifically the MATH and HiddenMath tests. Sujan Abraham, a senior software engineer at Labelbox, highlighted the versatility of these models, stating, “These models are designed for a wide range of tasks, including text, code, and multimodal applications. They can process larger and much more complex inputs like 1,000-page PDFs, massive code repositories, and hour-long videos.”
Furthermore, the models have shown improvements of ~2-7% in vision and code use cases. As a result, developers can expect better, faster, and more cost-efficient production environments.
Price Reductions and Increased Rate Limits
In a strategic move to enhance accessibility, Google has reduced the pricing for its Gemini 1.5 Pro model by more than 50% for both input and output tokens on prompts under 128,000 tokens. Specifically, the reductions are as follows:
64% price cut on input tokens
52% price cut on output tokens
64% price cut on incremental cached tokens
Additionally, the rate limits have been significantly increased, allowing developers to engage in more complex AI applications. The paid tier rate limits for the 1.5 Flash model have doubled to 2,000 requests per minute (RPM), while the limit for the 1.5 Pro model has tripled to 1,000 RPM.
Enhanced Speed and Efficiency
Developers can also expect improved speed and reduced latency with the new models. Google reported that the updated models offer 2x faster output and 3x lower latency compared to their predecessors. Jorge Argota, founder of an AI consultancy, remarked, “The Gemini 1.5 series is more efficient across the board. These models are text, code, and multimodal, making them more understanding and accurate when dealing with complex math or code. This could be a game changer for eCommerce platforms looking to implement advanced AI features.”
Flexible Content Moderation
Google has also made adjustments to the models’ default filter settings. Developers can now opt to use content safety filters based on their specific requirements rather than having them applied automatically. This flexibility allows for tailored configurations that better suit diverse use cases.
Conclusion
With these significant upgrades and reductions in pricing, Google aims to enhance the appeal of its Gemini platform to developers, especially in competitive sectors like eCommerce and retail. As the landscape of AI continues to evolve, Google’s efforts could catalyse broader adoption of its technologies across various industries.
