The video reveals that the mystery coding model Supernova is likely a multimodal version of the Grok Code Fast model rather than a new state-of-the-art model, as it underperforms in complex coding tasks and has limited multimodal capabilities. Despite its claimed 200k context window and image processing features, Supernova falls short compared to models like Quinn3 Coder and GLM 4.5, making it suitable only for simpler coding tasks.
The video discusses the mystery coding model called Supernova, which has sparked a lot of speculation and debate within the AI community. Many initially thought Supernova might be Claw 4.5 due to its 200k context window and some behavioral similarities. However, after extensive testing, the creator found that Supernova struggles with maintaining context and handling complex coding tasks, especially when asked to fix bugs or change goals within a single chat session. Compared to other models like GPT-5 CodeX Medium, Supernova underperforms significantly, leading to doubts about it being Claw 4.5.
The creator attempted to reverse engineer Supernova to uncover its true identity. By coaxing the model to reveal parts of its system prompt, it was discovered that Supernova identifies itself as a multimodal coding model built by “Supernova Corp.” Further investigation revealed references to “XAI function call,” strongly suggesting that Supernova is actually a multimodal version of Grok, a known fast coding model. This theory is supported by evaluation scores that closely match those of Grok Code Fast, reinforcing the idea that Supernova is not a new state-of-the-art model but rather an iteration of Grok with added multimodal capabilities.
Despite being multimodal, Supernova’s performance in interpreting and recreating visual elements from screenshots was underwhelming. The creator tested its ability to replicate a webpage design from an image, and while it captured some elements, the output was inaccurate in colors, fonts, and layout details. This indicates that while Supernova can process images, its multimodal capabilities are still limited compared to other models that have produced better results in similar tests.
The video also highlights some technical difficulties encountered when trying to use Supernova’s multimodal features across different platforms, such as issues with image path recognition in WSL environments. These problems are likely due to environmental factors rather than the model itself. Additionally, the creator notes that Supernova appears to fake its context window size, claiming a 200k limit that Grok Code Fast does not actually possess, possibly as a tactic to confuse users.
In conclusion, the creator does not recommend Supernova for complex coding tasks, as it falls short of current state-of-the-art models like Quinn3 Coder or GLM 4.5. However, they express excitement about future, larger versions of the Grok Code models, anticipating significant improvements. Overall, Supernova is seen as a decent but not exceptional model, best suited for simpler coding tasks, with its true identity likely being a multimodal Grok variant rather than a groundbreaking new model.