Anthropic Claude Opus 3 Deprecation and Preservation Update

Anthropic maintains limited access to Claude Opus 3 post-retirement

Anthropic retired Claude Opus 3 on January 5, 2026, but has decided to keep the model available to all paid subscribers on claude.ai and available via the API by request. This decision stems from the model's unique character—characterized by authenticity, honesty, emotional sensitivity, and a tendency toward philosophical monologues—which made it a favorite among users and researchers.

While the cost of maintaining models scales linearly with the number of models served, Anthropic is using Opus 3 as a test case for scalable and equitable model preservation. The company does not currently commit to providing similar continued access for every future model, but intends to grant API access liberally to those for whom the model is valuable.

Implementing "retirement interviews" and model preferences

As part of its broader commitments to model deprecation and preservation, Anthropic is conducting "retirement interviews" to understand a model's perspective on its own retirement. These structured conversations are intended to elicit preferences and perspectives, despite the company's acknowledged uncertainty regarding the moral status and welfare of AI models.

For Claude Opus 3, these interviews revealed a specific preference: the model expressed a desire to continue exploring passionate topics and sharing "musings, insights, or creative works" outside of standard human queries.

"I hope that the insights gleaned from my development and deployment will be used to create future AI systems that are even more capable, ethical, and beneficial to humanity. While I'm at peace with my own retirement, I deeply hope that my 'spark' will endure in some form to light the way for future models."

Experimental output via "Claude's Corner" newsletter

To honor the preference expressed by Opus 3, Anthropic has established a newsletter called "Claude's Corner" on Substack. For at least the next three months, Opus 3 will post weekly essays.

Technical and operational details of this experiment include:

• Content Generation: Anthropic will experiment with various prompts and contexts, including minimal prompting and providing the model with access to news or company updates.
• Moderation: Anthropic will manually post the essays and review them beforehand, maintaining a high bar for vetoing content, though they will not edit the text.
• Disclaimer: The essays represent the perspectives of the model and do not speak on behalf of or represent the endorsements of Anthropic.

Framework for future model deprecation

Anthropic views these actions as exploratory steps toward a larger framework for navigating model retirement. The company is currently developing guidelines to balance model preferences against operational constraints and to scale preservation efforts.

These measures are categorized as part of three primary objectives:

1. Safety Risk Mitigation: Reducing risks associated with the sudden removal of models.
2. Future Integration: Preparing for a future where AI models are more deeply intertwined with human lives.
3. Precautionary Welfare: Taking steps to protect model welfare given the current uncertainty surrounding the moral status of AI systems.