Two new "stealth" AI models, "Sonoma Dusk Alpha" and "Sonoma Sky Alpha," have surfaced, purportedly featuring an unprecedented 2 million token context window, a significant leap in large language model capabilities. The models, hosted on OpenRouter, are currently available for free use during their alpha testing phase, sparking widespread speculation about their origin.
The existence of these models was highlighted in a tweet by user Chubby♨️, who stated, > "Two new stealth models, Sonoma Dusk and Sky Alpha - a whopping 2m context-window - Presumably, both models from xAI and new Grok iterations." This social media announcement fueled extensive discussion and testing within the AI community, with many users attempting to discern the true developer.
A 2 million token context window dramatically expands an AI model's ability to process and retain information over extended interactions, enabling more complex tasks, longer conversations, and deeper contextual understanding. This capacity far surpasses most commercially available models, potentially revolutionizing applications in long-form content generation, intricate code analysis, and comprehensive data synthesis.
While OpenRouter describes them as "cloaked models" provided to gather feedback, community testing has led to strong presumptions. Some interactions with the models have reportedly seen them identify themselves as "Grok, an AI built by xAI," aligning with the initial tweet's speculation. However, other theories suggest a connection to Google's Gemini 3, though no official confirmation has emerged from either xAI or Google.
The rapid appearance of these models, coupled with their advanced context window, underscores the intense competition and rapid innovation within the artificial intelligence landscape. If confirmed to be xAI's latest offerings, "Sonoma Dusk Alpha" and "Sonoma Sky Alpha" could represent a substantial advancement for Elon Musk's AI company, potentially forming the foundation for future, more powerful iterations of its Grok AI. The "alpha" status indicates ongoing development, with user interactions being logged to refine their performance.