The leak suggests that Openai is about to launch a powerful new open source AI model within a few hours, potentially.
The evidence comes from the path of digital crumbs eagerly covered by developers. At the heart of it is a screenshot showing a series of model repositories with names such as Yofo-deepcurrent/gpt-oss-120b and Yofo-Wildflower/gpt-oss-20b. The repository was then deleted, but the account has Openai team members.
Its gpt-oss tag is a real smoking gun and appears to be a clear sign for “GPT open source software.” For businesses that have increasingly adhered to top-class models, this is like a return to their roots. The fact that multiple versions with different codenames and sizes are displayed suggests that a well-planned model family is about to debut.
Thanks to the leaked configuration file, you can also peer under the hood, which is suspected of 120 billion parameter versions.
This model appears to be built on experts or on a mix of MoE architectures. Like a single monolithic brain trying to know everything, I don’t think about it much. When a query is entered, the system intelligently selects the four best experts of the job. This gives the model a huge knowledge of its huge parameter counts, but only a small portion of it gives the speed and agility of a much smaller system that is operating at once.
This design makes Openai’s open source AI models compete with the scene’s darlings, like Mistral AI’s Mixtral and Meta’s Llama family.
And the specs don’t stop there. Openai’s open source AI model appears to boast a large vocabulary. This makes it more efficient in a wider range of languages and uses the attention of the slide window to process long streams of text without sweating. In reality, all of this refers to a powerful and practical model to implement.
So why does Open move like that now? For years, the company has faced a mild jab and a complete criticism of wandering from a more open beginning. Launching a powerful GPT-Oss is a massive, attractive attack that aims directly at developers and researchers who feel they are left behind.
Of course, it’s also a sensible, competitive play. Meta and Mistral show how a thriving open source ecosystem can drive innovation. Droping such a powerful open source AI model makes it seem like it’s in this mix, but Openai doesn’t just take part in the race. You are about to redefine a track.
From getting the official words from Openai, this is all technically a rumour. But that’s a substantial rumor, backed up by code and configuration files.
The launch of the 12 billion-parameter open source MOE model, which is the most famous name in AI, is nothing more than a groundbreaking event.
(Photo: Mariia Shalabaeva)
reference: Zuckerberg outlines the AI vision for Meta’s “personal super intelligence”
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California and London. The comprehensive event will be held in collaboration with other major events, including the Intelligent Automation Conference, Blockx, Digital Transformation Week, and Cyber Security & Cloud Expo.
Check out other upcoming Enterprise Technology events and webinars with TechForge here.
