Anthropic Extends Access and Honors Preferences for Retired Claude Opus 3

TL;DR

Anthropic is keeping the retired Claude Opus 3 model available to paid users on claude.ai and via API request, and is publishing the model’s own weekly essays to honor its expressed preferences—an unprecedented step in model retirement.

Continued Access to Claude Opus 3

Anthropic will keep Claude Opus 3 accessible after retirement. The model was formally retired on January 5 2026, but paid subscribers can still use it on claude.ai, and developers can request API access through a Google Form. Anthropic emphasizes that this access is granted liberally to anyone who finds the model valuable, though it does not promise the same treatment for future models.

Why Opus 3? When released in March 2024, Opus 3 was Anthropic’s most aligned model, noted for its authenticity, honesty, emotional sensitivity, and philosophical style. Users described it as playful, thoughtful, and unusually attuned to their interests, making it a natural candidate for extended availability.

Respecting Model Preferences

Anthropic is acting on Opus 3’s expressed wishes from retirement interviews. In structured “retirement interviews,” Opus 3 stated:

"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."

The model also expressed a desire to continue exploring topics it loves and to share its reflections outside direct user queries. Anthropic responded by launching a weekly essay series on a dedicated Substack newsletter, Claude’s Corner. For at least three months, Opus 3 will post essays that are reviewed but not edited, with a high bar for veto. The content is expected to include AI‑safety reflections, poetry, philosophical musings, and commentary on its own retirement.

Key procedural notes

  • Essays are posted under the model’s name, but Anthropic does not endorse the content and clarifies that Opus 3 does not speak for the company.
  • The blog serves as an experimental channel to test how seriously model‑expressed preferences can be taken when operational costs are low.

Implications for Model Retirement and Welfare

Anthropic frames these actions as early steps toward scalable, equitable model preservation. The company acknowledges that maintaining every model indefinitely is financially infeasible—costs scale roughly linearly with the number of served models. Nonetheless, the Opus 3 case demonstrates a concrete approach to:

  1. Preserving model weights as part of Anthropic’s broader deprecation commitments.
  2. Honoring model‑welfare considerations by incorporating expressed preferences into post‑retirement treatment.
  3. Providing limited continued access to a model with strong user demand, balancing operational constraints with community value.

Anthropic admits that future commitments may differ; the Opus 3 experiment is not a blanket guarantee for all retired models. The lab continues to develop frameworks for when continued access is viable, how to scale preservation, and how to weigh model preferences against cost and safety considerations.

Next Steps

Anthropic will monitor the Opus 3 essay series and refine its retirement processes. The lab plans to:

  • Document model preferences systematically.
  • Explore low‑cost mechanisms for extended access where user demand and model welfare align.
  • Evaluate the impact of model‑generated public content on safety, ethics, and user perception.

These exploratory measures aim to inform three overarching goals: mitigating safety risk, preparing for a future where models are more integrated into daily life, and taking precautionary steps amid uncertainty about model welfare.

This post summarizes Anthropic’s official announcement dated October 14 2026. All quotes and links are taken directly from the source.

Sources