Google on Tuesday released new versions of the Gemini 1.5 Pro artificial intelligence (AI) models. These come just months after the Mountain View-based tech giant released the final version of Gemini, which increased the reference window to 2 million tokens. Dubbed Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002, the company said that these AI models not only offer high output and low cost, but will also provide users with a higher rate limit. Additionally, filter settings have also been updated to make the AI model follow instructions more closely.
New Gemini AI models released
In a blog post, the company detailed about the Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002 AI models. These models are currently available as experimental model releases and are built on top of Gemini 1.5 Pro, which was first released at Google I/O in May. Currently, these are available to the company’s developers and enterprise customers. Developers can access it for free from Google AI Studio and Gemini API. Enterprises can access it through Vertex AI.
According to internal testing conducted by Google, the latest Gemini 1.5 Pro and Flash models even outperformed the previous generation Gemini models. The company claimed that the new models have shown a seven percent increase in the Massive Multitask Language Understanding Pro (MMLU-Pro) benchmark. Additionally, the AI models are said to offer approximately 20 percent improvement on MATH and HiddenMath benchmarks compared to Gemini 1.5 Pro.
The GEMINI-1.5-PRO-002 and GEMINI-1.5-FLASH-002 AI models will also offer increased rate limits. Rate limits are daily usage limits for users. With the 1.5 Flash model, users will get 2,000 requests per minute (RPM) and the 1.5 Pro model will offer 1,000 RPM. Google said these limits are being increased to allow developers to build with newer versions of Gemini.
It’s not just the rate caps that are being upgraded. With gemini-1.5-pro-002 and flash-002, the company has also increased the output tokens per second, making the models more responsive and faster to generate longer blocks of text.
One major upgrade with these AI models is the improvements to the filters. Google said the new Gemini AI models will better follow signals and follow instructions because of these updated filters. Google is also improving its security measures to ensure that AI models are not generating anything harmful. Specifically, the new AI models will not have default filters applied so developers can choose their preferred configuration.
