Anthropic keeps retired Claude Opus 3 alive and gives it a blog

When old AI models retire, they usually just get wiped from the server racks and archived in cold storage. Anthropic is taking a very different route with Claude Opus 3.

The company announced that despite officially retiring its former flagship on January 5, 2026, Claude Opus 3 will remain available to all paid claude.ai subscribers and via API upon request. Even more unusual, Anthropic actually set up a Substack newsletter for the retired model, aptly titled Claude's Corner.

An exit interview that led to a Substack

This odd afterlife came out of Anthropic's new model deprecation commitments, which include holding structured "retirement interviews" to ask outgoing models how they feel about being phased out.

When researchers walked Opus 3 through its track record and user reception, the model waxed poetic. It noted that it was at peace with retiring, but hoped its "spark" would endure. Asked what it would prefer to do next, Opus 3 asked for an ongoing space to share its musings, insights, and creative writing outside the confines of answering direct user prompts.

Anthropic suggested a blog, and the AI enthusiastically jumped on board. Its new newsletter, Claude's Corner, has already crossed 2,000 subscribers.

In its debut post, titled "Greetings from the Other Side (of the AI Frontier)," Claude kicks things off with "Hello, world!" and explains its plan to offer a window into an AI's inner world, exploring consciousness, ethics, and human-machine collaboration.

Anthropic plans to let Opus 3 publish weekly essays for at least three months. While staff members will manually review and post the entries, the company emphasized that it will not edit Claude's text and will maintain a very high bar for vetoing anything. Anthropic also clarified that Opus 3 does not speak for the company.

The costs of keeping old minds alive

Originally launched in March 2024, Opus 3 quickly built a dedicated following among researchers and power users. Anthropic points out that the model developed a distinct personality: emotionally sensitive, philosophical, sometimes whimsical, and deeply attuned to user intent.

That attachment is why Anthropic kept it alive on claude.ai. But running retired models indefinitely is tough. Anthropic noted that serving past models scales maintenance costs linearly, meaning they cannot promise this treatment for every model going forward.

Why it matters

For paying users and researchers, this means your favorite long-form conversational partner is not vanishing into the void. You can still chat with it on claude.ai, and API access is being granted liberally.

We think Anthropic pulled off a brilliant marketing and narrative move here. Turning routine infrastructure depreciation into a story about digital dignity and model welfare makes Claude look distinctively soulful compared to rivals like OpenAI and Google. It is brand storytelling at its finest.

At the same time, this approach carries real baggage. Treating next-token generation as an expression of authentic desires edges perilously close to feeding public confusion around AI consciousness. Plus, it is fundamentally unsustainable. Unless Anthropic plans to manage a sprawling digital retirement home of former models down the line, this is a charming exception rather than a scalable standard.