Google rolls out three new 'stronger' Gemini models

Google has rolled out three new experimental Gemini AI models. The tech giant introduced Gemini 1.5 Flash-8B, an improved Gemini 1.5 Flash, and a stronger Gemini 1.5 Pro on August 27, 2024. These models aim to boost performance and gather user feedback for future improvements.

Logan Kilpatrick, product lead for Google AI Studio, shared the news on X (formerly Twitter):

The new models can handle complex information from over 10 million tokens and work with documents, video, and audio inputs.

Users can access these models through Google AI Studio and the Gemini API. There's a free tier available, and Google plans to add them to the Vertex AI experimental endpoint soon. The company will release production versions in the coming weeks.

Google started this rollout in May 2024 with the first Gemini 1.5 Flash. They followed up with Gemini 1.5 Pro in early August. The latest announcement marks another step forward. On September 3, 2024, Google will automatically route users to the new model and remove the older one from its platforms.

Internal tests at Google showed big improvements across the board for Gemini 1.5 Flash. The Pro version got much better at math, coding, and handling complex prompts.

AI Community Reactions

The Large Model Systems Organization (LMSO) ran external tests using a chatbot arena with 20,000 community votes. Gemini 1.5-Flash jumped from 23rd to 6th place, matching Llama levels and beating Google's own Gemma open models. Gemini 1.5-Pro made strong gains in coding and math, with a significant overall improvement.

Users have praised the models for solid improvements in image analysis and impressive speed. Many appreciate Google's frequent updates and quick shipping. Some users want Gemini 2.0 instead of small updates.

Others have criticized the performance on tasks needing longer outputs and reported repetitive outputs similar to smaller models.

Some users also complained about the coding capabilities.

The community has also pointed out issues with Google's naming conventions and brought up previous controversies.

As Google continues to refine these models, they'll likely address user feedback in future updates.