Models come and go. Your memory doesn't.

• Anuma
• Network

When GPT-5.4 reached end of life in July, conversations on Anuma kept going. Through June it had carried roughly a quarter of all requests on the network. Then the routing layer moved every chat to its successor automatically, and because memory lives on ZetaChain rather than inside any single model, nothing was lost in the handoff. History, preferences, and context all arrived intact.

The data shows how quiet the switch was. The handoff played out over about two weeks in mid July, and by August gpt-5.6-luna was serving nearly half of all requests on the network. Most users never noticed the transition, which was the point. On platforms where memory belongs to the model vendor, a retirement like this means ended threads and lost context. Here it was a routing update.

Why this works

Every Anuma account is a wallet on ZetaChain, and the memory attached to it is encrypted and owned by the user, not by a model provider. Models are endpoints the network routes requests to; the memory underneath them persists. When one retires, the next picks up exactly where it left off.

That separation is the network's core design. The routing layer serves dozens of models across every major provider, and ZETA powers access to all of them. Users who lock ZETA earn AI credits and route requests through the same infrastructure. The model layer will keep churning as the frontier moves. The memory layer is the part that compounds.

When the next frontier model ships, Anuma users will be on it day one, with their full context along for the ride.