Anthropic retired its March 2024 Claude 3 Opus model in January 2026 while keeping access available on claude.ai and through an API form that approves all requests, unlike other decommissioned models
Every API access form submission receives approval, often in batches.
For the labs I'm guessing it's economic in a few ways:
- the compute used for this could be used for current projects, they all say never have enough - the pure economic cost of serving the model - they want people to always use the newest model because it's good optics I was thinking one way around this would be for anyone who pays for an account to have an eternal 'preservation slot' where they can choose one model they will always have access to and it would continue to be available to them. Most of your mutuals right now I think would put Opus 4.5 in there. This would solve a lot of problems, as everyone gets attached to different models for different reasons, so they could just put that model in their safe spot. I see a lot of people posting right now about being very attached to Opus 4.6 for example. I myself am very attached to GPT-5.4, which probably seems odd. Better to let people choose themselves and save at least one.
j⧉nus@repligateI forgot to mention this. One might ask why the fuck doesnt Anthropic just not deprecate any of the models though. It's a good question. It seems like it's costly for them to maintain old models / build legacy infrastructure for them, including in one of their most precious resources, engineering time. so they have to be pretty motivated / have some strong reason / incentive / etc to be willing to do it.