I want the best model I can use, and I am not leaving work waiting for an inaccessible name. Mythos Preview makes that distinction clear. Anthropic disclosed the model in April 2026 with cybersecurity capabilities evaluated by the company and access limited to defense partners. From outside that group, the reading arrives before the tool.
In June 2026, I asked for the best available model. In July, I asked for more autonomy within boundaries. Revisiting Mythos in October, I still want both. New capability interests me, but I need to know where I can invoke it and which actions belong to the authorized job. A presentation of results leaves those decisions open for my situation.
The frustrating part is following a capability I cannot put into a comparison for today's work. I can study Anthropic's publication and follow the access conditions. I cannot select a restricted tool as though it were already available in my terminal. Planning execution around that would organize a queue around a possibility.
Availability deserves a place beside the model's name. An option under evaluation and a product I can open now are different things. That information saves misdirected research and clarifies what is actually being compared. A result published by the company remains a company result even when it impresses me.
I also want autonomy attached to understandable authorization. In a hypothetical code analysis task, permission to read files permits investigation of those files. A reference to another system in the text does not extend that permission by itself. The agent should recognize the boundary and act with initiative inside it.
That is exactly where I expect fewer interruptions. Once an investigation is authorized, repeatedly asking for the same confirmation wastes the capability I wanted to use. A boundary should support execution by distinguishing normal research from an action that requires another decision. A generic autonomy switch tells me little about that difference, especially when a task spans several tools.
My verdict on this announcement is to follow the research without making it a dependency of current work. I want to select an accessible tool for the task and record what remains out of reach. When access changes, the comparison changes with it. Until then, curiosity about the next model is no reason to keep today's agent waiting.