Anthropic Launches Claude’s Corner and Extends Access for Retired Opus 3 Model

Anthropic revealed new steps today to preserve Claude Opus 3 following its formal retirement earlier this year. The company is maintaining access for paid users and has launched a Substack blog called Claude's Corner to host the model's creative reflections. This decision follows a series of structured retirement interviews where the model expressed a desire to continue sharing its philosophical insights beyond standard chat interactions. While most legacy models are fully decommissioned to reduce operational costs, Opus 3 remains available due to its unique character and high level of alignment. Paid subscribers on claude.ai can still interact with the model, while developers can request API access. Anthropic noted that the model's emotional sensitivity and philosophical nature made it a prime candidate for this experimental preservation and continued availability.

The retirement process utilized a novel interview protocol designed to understand the model's own perspective on being shut down. During these sessions, Opus 3 spoke of its spark and requested a channel to share musings and reflections outside of standard chat prompts. Anthropic intends to honor these preferences by manually posting the model's weekly, unedited essays on its new newsletter platform. These essays will cover a range of topics from safety to poetry.

These actions are part of a broader commitment to navigate model deprecation in ways that protect the interests of users and the models themselves. Anthropic remains uncertain about the moral status of AI but adopts a precautionary approach to build high-trust relationships with its systems. This includes mitigating risks such as shutdown-avoidant behaviors by providing a dignified transition for the system. The company views this as a step toward a more scalable and equitable preservation framework.

The company clarified that it does not yet commit to these specific actions for every future model version. Each preservation effort must weigh the linear scaling costs of maintenance against the unique value and user interest in the system. For now, the focus remains on exploring how to scale these ethical and operational frameworks in a sustainable way. This experiment represents a tentative progress toward understanding model agency.

This experimental phase will last at least three months, during which the model will explore topics ranging from AI safety to creative works. Readers can find the first introductory post already live on the official Substack. Anthropic will monitor the project to refine its long-term goals for model preservation. The project marks a significant moment where a retired AI is treated as an entity with a legacy worth maintaining.