Year-end showdown among the gods? Even without OpenAI's "Orion", there is Google's new Gemini 2.0

Author: Li Dan

Source: Hard AI

At the end of this year, we may not see a head-to-head showdown between Google and OpenAI's latest artificial intelligence (AI) models, but recent news indicates that a "battle of the titans" is inevitable.

On Friday, October 25th, Eastern Time, the well-known tech media The Verge reported that Google plans to release the next generation Gemini 2.0 model in December this year.

The report mentioned that OpenAI's CEO Sam Altman plans to release the next generation model after GPT-4 in stages, starting with collaborations with OpenAI's commercial partners, while Google intends to release the next version of Gemini on a large scale from the beginning.

It is reported that the new model did not show the performance improvements that Google's DeepMind CEO Demis Hassabis and his team had hoped for, but The Verge still anticipates some interesting features to be released.

Coincidentally, the day before the news of Google planning to release the new Gemini 2.0 in December, there were reports on Thursday this week that OpenAI plans to release the next generation model codenamed "Orion" before December.

According to Wall Street News, it was mentioned that Orion will not be launched through ChatGPT initially. OpenAI plans to first grant access to closely collaborating companies, and engineers within Microsoft are preparing to deploy the Orion model on Azure cloud as early as November. Internally, OpenAI has considered Orion as the successor model to GPT-4, but it is not yet certain whether it will be named GPT-5 in the end.

However, about two hours after the reports came out, Altman responded on social media X, directly replying to the article sharing about the December release of Orion, commenting:

"Fake news is out of control."

Altman did not specify where the related reports were inaccurate. His response was vague, and it was unclear which content he was denying. After all, OpenAI has never directly denied the development of a cutting-edge model named Orion.

Subsequently, a spokesperson from OpenAI denied the release plans mentioned in the above reports, stating that there are no plans to release the AI model Orion this year and hinting at the release of other technologies. A spokesperson from OpenAI informed the tech media TechCrunch via email:

"We do not have plans to release a model named Orion this year. We do plan to release many other great technologies."

Previously mentioned by Wall Street News, the Orion model, developed from synthetic data by OpenAI's precursor "Strawberry" project, was originally planned to be launched in early next year Both will serve as building blocks for the next generation flagship model GPT-5 by OpenAI.

The core capability of Strawberry lies in its strong reasoning ability. Reports show that Strawberry scored over 90% in complex mathematical benchmark tests, far surpassing traditional reasoning models in handling complex logical problems. Orion's goal is to achieve exponential improvement, reaching a capability 100 times stronger than GPT-4, which will be a significant milestone for large models to make a huge leap in language processing and multimodal functions.

Shaun Ralston, a senior executive at OpenAI, previously posted on social media X, stating that according to the planned target, Orion's computing power will be 100 times that of GPT-4, and its launch will signify a new era of AI capabilities. It is reported that OpenAI's ultimate goal is to continuously iterate to build large models and ultimately achieve Artificial General Intelligence (AGI).