What Phase 1 does not include
Call /v1/models today and you get eight entries back. Two of them will answer a request. One is on canary routing. The other five are registered, specified, and dark.
This is deliberate, and it is the single design decision on the API most likely to be read as a mistake. A catalogue that lists models you cannot call looks like a catalogue with bugs in it. So here is the reasoning.
A dark surface is not an outage
Every surface in the codex has a phase that activates it. ogi-reasoning-1 activates with Phase 3; ogi-manipulation-1 with Phase 4. Requesting one today returns a 400 with the code model_not_active, and the error body names the phase and the status rather than saying "unavailable":
{
"error": {
"message": "`ogi-reasoning-1` is specified but dark. It is registered in the
catalogue and activates with phase 3; no validator is serving it today.",
"type": "invalid_request_error",
"code": "model_not_active",
"param": "model",
"status": "planned",
"activates_with_phase": 3
}
}The alternative was to omit dark surfaces from the catalogue entirely and add them as they activate. That reads better and tells you less. It would imply that what is currently serving is the whole of what has been specified, which is false, and it would make each activation look like new capability rather than a scheduled one.
What Phase 1 is actually gated on
The roadmap defines phases by milestone rather than by calendar. Phase 1 exits on sustained inference throughput, audit divergence below the published threshold, and a published count of independent validators across at least two jurisdictions. No date appears anywhere in that sentence, on purpose — a phase ends when its criteria are met, and a date on a roadmap is a promise about a thing the people making it do not control.
The consequence is that Phase 1 can sit where it is for a long time, and the network will keep saying so through /v1/network/status rather than quietly moving the marker.
The one that matters most is last
Worth saying plainly, because it is the part most likely to be skipped: the final phase is not a capability. Phase 6 is operator sunset, and its exit criterion is that no foundation-distinguished role remains active. Multisig keyholders rotate on a published schedule, foundation-operated services hit their sunset clauses, and parameters that the foundation decided move to governance.
P11 calls this the most aggressive design choice in the codex, and it is: permanent operators are the failure mode every previous decentralized system has eventually exhibited. The response is to make the operator role temporary by construction rather than by promise. Everything before Phase 6 is capability. Phase 6 is the point.