Google has recently made another significant push in the AI field, launching the latest Gemini1.5 model. This version includes a smaller variant, Gemini1.5Flash-8B, as well as "significantly improved" Gemini1.5Flash and "more powerful" Gemini1.5Pro.
According to Google, compared to many internal benchmark tests, performance has improved, with the overall performance of Gemini1.5Flash experiencing a "massive boost," while the 1.5Pro is much better at mathematics, coding, and complex prompts.
Logan Kilpatrick, Google's AI product manager, stated on social media: "Gemini1.5Flash is now the best choice for developers worldwide!" The Gemini1.5 series models can handle long texts and perform reasoning over more than 10 million tokens, enabling them to process large volumes of multimodal inputs such as documents, videos, and audio.
Latest Experimental Iteration of the Gemini Model
Google introduced the lightweight version of Gemini1.5, Gemini1.5Flash, in May. The Gemini1.5 series models are designed to handle long contexts and can reason over fine-grained information from 10 million and more tokens. This allows the models to process large-volume multimodal inputs, including documents, videos, and audio.
Google has now released a smaller version of Gemini1.5Flash with 8 billion parameters. The new Gemini1.5Pro has seen significant improvements in performance for coding and handling complex prompts. Kilpatrick said that Google will release a production-ready version in the coming weeks and hopes to provide more evaluation tools.
According to Kilpatrick, the release of these experimental models is to gather feedback so that the latest updates can be delivered to developers faster. He revealed that the new models will be provided for free to developers, tested through Google AI Studio and the Gemini API, and will also be released through experimental endpoints of Vertex AI in the future.
Just a few hours after the release, the Large Model System Organization (LMSO) updated its chatbot rankings based on 20,000 community votes. Gemini1.5-Flash made a "huge leap," climbing from 23rd to 6th place, on par with Llama, outperforming Google's Gemma open model.
Starting from September 3, Google will automatically direct requests to the new model and remove the old versions to avoid confusion. Kilpatrick is full of anticipation for this new model, hoping to see developers leverage it for more multimodal applications.
However, with the release of the new model, early feedback has shown a polarized trend. Some users are puzzled by the frequent updates, believing that users are still looking forward to the more comprehensive Gemini2.0 version. At the same time, some users praise the speed and performance of the updates, believing that this keeps Google at the forefront of the AI field.
Key Points:
✨ The newly released Gemini1.5 series models have shown significant improvements in mathematics, programming, and complex prompts.
🚀 Gemini1.5Flash can handle multimodal inputs over 10 million tokens, with strong adaptability.
🛠️ User feedback is polarized, with some looking forward to the comprehensive upgrade of Gemini2.0, while others appreciate the current updates.