Google is making significant updates to its Gemini models as it moves closer to the release of the 2.0 version. The company has announced the launch of three experimental models: Gemini 1.5 Flash-8B, Gemini 1.5 Flash, and Gemini 1.5 Pro. These models show improved performance in various areas, such as math, coding, and complex prompts. Logan Kilpatrick, product lead for Google AI Studio, claims that Gemini 1.5 Flash is currently the best model for developers.
Gemini 1.5 Flash is a lightweight version of Gemini 1.5, designed to handle long contexts and process high-volume multimodal inputs. Google is now releasing an improved version of a smaller 8 billion parameter variant of Gemini 1.5 Flash. The new Gemini 1.5 Pro offers performance gains in coding and complex prompts and serves as a replacement for the previous model released in August. Google plans to release a future version for production use in the coming weeks.
These experimental models are an opportunity for Google to gather feedback and provide developers with the latest updates as quickly as possible. Kilpatrick explains that the insights gained from these launches will inform wider model releases. Both Gemini 1.5 Flash and Pro are available for free testing through Google AI Studio and Gemini API.
Starting from September 3, Google will automatically reroute requests to the new models and remove the older models to avoid confusion. The Google DeepMind researchers describe Gemini 1.5’s scale as “unprecedented” among contemporary LLMs. Kilpatrick expresses excitement about the feedback and the potential for these models to unlock new multimodal use cases.
Shortly after the release, the Large Model Systems Organization (LMSO) updates its chatbot arena leaderboard based on community votes. Gemini 1.5-Flash makes a significant leap in the rankings, climbing to sixth place and outperforming Google’s Gemma open models. Gemini 1.5-Pro also shows strong gains in coding and math. The LMSO congratulates Google DeepMind’s Gemini team on the successful launch.
Feedback on the new models has been mixed, ranging from praise to criticism. Some users question the frequent updates and suggest the need for a 2.0 version. However, many users appreciate the fast upgrades and commend Google for continuously delivering improvements. Some critics argue that the models still have limitations and lag behind competitors like Claude, OpenAI, and Anthropic.
There is also some mockery of Google’s naming choices and a reference to a controversial incident earlier this year. However, overall, there is anticipation among users to try out the new models and see the improvements in action.