Skip to content
Source
Sign inSign up
TanStack/schemas

Backdate first seen #1

Send to my agent
Closedtombeckenham opened this issue 0 comments
tombeckenhamcommented

Backdate models.firstSeenAt to real per-provider release dates so the catalog shows historical "first seen" dates instead of when our poller happened to first observe each model.

Agreed approach (2026-07-14)

Semantics: backdate the existing firstSeenAt column in place — move it earlier when the provider reports a release date before our observed value, never later. No new column, no API change.

Sources per provider (verified against live APIs):

ProviderSource of release date
OpenAIcreated (epoch seconds) in GET /v1/models — 100% coverage
Anthropiccreated_at (ISO 8601) in GET /v1/models
xAI Grokcreated (epoch seconds) in GET /v1/models
OpenRoutercreated (epoch seconds) in GET /api/v1/models — 100% coverage
FALmetadata.date (ISO 8601) in GET /v1/models
Geminino API field — curated date map, plus parsing MM-YYYY suffixes embedded in preview ids (e.g. gemini-2.5-computer-use-preview-10-2025)
ElevenLabsno API field — curated date map

Mechanism: continuous via the existing 15-min models poll (not a one-time backfill). ModelInfo gains an optional releasedAt (epoch seconds); each provider's listModels populates it; poll-models.ts uses it on insert (firstSeenAt = min(now, releasedAt)) and backdates existing rows whenever upstream reports an earlier date. Existing prod rows self-heal on the next cron run — no admin endpoint or manual step.

Models with no known date (unmapped Gemini/ElevenLabs entries) keep poll-time firstSeenAt; the curated map is easy to extend later.

  1. tombeckenham closed this

Sign in to join this conversation.