Google launched two stable versions of its Gemini 1.5 application programming interface models, offering developers improved performance and reduced costs for app production.
Gemini 1.5 Pro and Flash
Google announced the release of Gemini 1.5 Pro and Gemini 1.5 Flash stable versions.
Compared to the previous 001 models, these production-ready Gemini models have shown huge progress in code generation, math, reasoning, and video analysis.
Cheaper but better
Google lowered the price of its Gemini 1.5 Pro model by more than 50% while claiming three times higher rate limits and lower latency than the older experimental model releases.
Today, we’re excited to release two new, production-ready versions of Gemini 1.5 Pro and Flash. 🚢
They build on our latest experimental releases and include significant improvements in long context understanding, vision and math. → https://t.co/C08ElHVE8e pic.twitter.com/DJsOsBwDmb
— Google DeepMind (@GoogleDeepMind) September 24, 2024
Both Gemini 1.5 models offer nice gains in factuality, reduced model hallucinations, improved instruction following, enhanced multilingual understanding in 102 languages, better SQL generation, and better audio and document understanding. Quite an update stack, isn’t?
Google also reduced the summarization lengths for both models and advised chat-based product developers to explore options for increasing the API’s conversational capabilities.
From October 1, Gemini 1.5 Pro API prices on prompts less than 128,000 tokens will be reduced to 64% for input tokens, 52% for output tokens, and 64% for incremental cached tokens.
Additionally, the paid tier rate limits for 1.5 Flash will be increased to 2,000 RPM, and 1.5 Pro will be increased to 1,000 RPM, up from 1,000 and 360, respectively.
Experimenting with the new models
Another big step is the launch of Gemini 1.5 Flash-8B, a smaller experimental version of 1.5 Flash with lower benchmark numbers.
This update includes huge performance increases across both text and multimodal use cases. All versions are currently available at Google AI Studio and the Gemini API.
Meanwhile, Google’s main artificial intelligence competitor, OpenAI, has begun rolling out its new Advanced Voice feature to select ChatGPT users.
Advanced Voice is rolling out to all Plus and Team users in the ChatGPT app over the course of the week.
While you’ve been patiently waiting, we’ve added Custom Instructions, Memory, five new voices, and improved accents.
It can also say “Sorry I’m late” in over 50 languages. pic.twitter.com/APOqqhXtDg
— OpenAI (@OpenAI) September 24, 2024
This new feature allows for faster and more intuitive human-like communication with AI, and includes five new voices: Arbor, Maple, SXol, Spruce, and Vale, which are additions to the existing Breeze, Juniper, Cove, and Ember voice options.
These improvements to Google’s Gemini 1.5 models and OpenAI’s ChatGPT voice feature could – and pretty likely will – lead to increased adoption and usage of these AI platforms both by developers and users.
