If you’ve built workflows around Claude Opus 3, this is your cue to pay attention.
Anthropic is officially “retiring” Claude Opus 3 as a standard, generally available model — and using the move to debut what it calls its first formal model retirement process. The reason is familiar: it’s costly and operationally complex to keep older models broadly available while newer ones ship.
This also isn’t a hard cutoff. Anthropic says Opus 3 will remain accessible on claude.ai for paid subscribers, and developers can still get it through the API by request. For teams that tuned systems around Opus 3’s specific behavior and “feel,” this is both a practical heads-up and an early look at how Anthropic may handle future sunsets.
Not the default, but not gone
According to Anthropic, supporting multiple AI models simultaneously incurs ongoing costs that scale “roughly linearly.”
In Opus 3’s case, the company is keeping an access path open because demand remains high and it considers the model worth studying. It also describes this as an early experiment.
For enterprise teams, that distinction matters. Even if Opus 3 remains reachable through subscriptions or by request, “retired” status can affect how it’s handled in procurement, governance reviews, and internal standards, especially once it’s no longer the default.
The ‘most aligned model to date’
Anthropic cast Opus 3 as both a fan favorite and a research-worthy outlier. It described a “constellation of traits” that made it beloved, citing an interaction style that includes sensitivity, playfulness, and a philosophical streak. The company also noted that when Opus 3 launched in March 2024, it was Anthropic’s “most aligned model to date.”
More broadly, Anthropic positioned this as a real-world test on how to preserve select models people care about while acknowledging real limits on what a provider can keep running and do it in a way that remains scalable and equitable.
Retirement interviews and Claude’s Corner
A distinctive element of the process is “retirement interviews,” structured conversations meant to capture a model’s perspective and, when possible, reflect its preferences. Anthropic added a caveat: the outputs can be shaped by context and trust, so these interviews aren’t a neutral readout.
For Opus 3, Anthropic said the model wanted a space for ideas beyond typical chat. The AI company proposed a blog-style outlet; it says Opus 3 agreed and asked for a place to publish “musings and reflections.” That became Claude’s Corner, where the model will publish weekly essays for at least three months.
Anthropic noted that it will review and manually post the essays without editing them, setting a “high bar” for vetoing content. It also emphasizes that the posts don’t represent Anthropic and may not align with the company’s views.
The bigger picture: deprecation becomes part of the product
Anthropic is trying to make model retirement feel like lifecycle management, with continued-access options and tradeoffs stated upfront. It acknowledges that deprecations can disrupt users who depend on a specific model, narrow what researchers can study, and raise safety and welfare questions. The company presented this approach as an attempt to balance those pressures.
The takeaway: model retirement is becoming routine. Vendors may increasingly differentiate not only on performance, but on how predictable their phase-out process is, what preservation routes exist, and how clearly they communicate access terms and timelines.
A $200 million AI standoff is unfolding inside the Pentagon as Anthropic’s CEO faces mounting pressure over military access to Claude.