Google is expected to release the next version of its Gemini family of AI models early next month, but it may come sooner than expected, according to Testing Catalog.

Gemini was first announced in December 2023, replacing the Bard chatbot and becoming the default model in all Google products. Since then, version 1.5 has been released and multiple small upgrades have been made, but Gemini 2 appears to be just around the corner.

Alexey Shabanov of Testing Catalog reports that he found a new “experimental model” listed under the model menu in the Gemini web app. It was under the Gemini Advanced option and was labeled “With 2.0-Pro-Exp-0111”. It is not clear if this is actually Gemini 2 or if it is mislabeled.

Shabanov's test reported that it was “lightning fast” but may not be “perfectly baked” because it failed several basic tests, including identifying the number of times the “r” appears in the word strawberry.

We know very little about Gemini 2. This is because AI labs, including those funded by major tech companies, tend to keep the details close to their chests until all training and testing is complete. But thanks to leaked information and early external testing that went wrong, we have some brief insight.

Gemini 2 will likely have inference capabilities similar to OpenAI's o1 model family. It will also power “Project Jarvis,” a new agent system that will be incorporated into Chrome that will be able to perform common web tasks using natural language prompts.

Earlier reports indicated that it would be released in the first week of December, one year after the release of Gemini 1, which makes sense and would be around the time when a major upgrade of ChatGPT (to coincide with ChatGPT's second anniversary) is expected.

If Gemini 2 follows Google's recent trend, there will initially be two versions, Gemini 2 Flash for very fast, low-level response and Gemini 2 Pro for more advanced output. If this leak is accurate, it appears that the speed will be upgraded in the Pro as well.