Skip to content
Source
Sign inSign up
TanStack/schemas

Reactor: /pricing renamed x2 to xmax/x2, so x2 loses its price every poll (and listModels hides fetch failures) #100005

Send to my agent
Closedtombeckenham opened this issue 1 comment
tombeckenhamcommented

What fails

From the 2026-11-17 21:52 UTC poll, Reactor emits pricing_lost (parser_miss, rawId x2) plus an update_storm (PostHog project 643425).

Why

GET https://api.reactor.inc/pricing has renamed the model. It used to list x2. It now lists xmax/x2 (id 75b9886b-1fb5-4bb3-873a-6a44f5b8c70d, 17 credits/s = $0.0017/s) and a new reactor/xmax-x2-1 (same rate).

REACTOR_MODELS pins { pricingName: 'x2', rawId: 'x2' } (src/server/providers/adapters/reactor.ts:107). MODEL_BY_PRICING therefore misses xmax/x2, and two things happen:

  • connectSlugFor turns xmax/x2 into a new row, xmax/x2.
  • the curated x2 row is appended with no rate, so the stored card is kept and pricing_lost fires.

The listing doesn't change between polls, so this repeats every poll.

Fail-loud gap in the same path

listModels (reactor.ts:652-666) catches every error and returns curatedCatalog(), and an empty or unreadable /pricing body does the same (rows.length === 0). A broken pricing fetch or parse then looks like a successful poll with no prices. It should throw so the poll reports ingest_failed, the same as other adapters' listings.

Fix

  1. Map the renamed model. Check Reactor's model page for whether xmax/x2 is the same model as the stored x2. If it is, set pricingName: 'xmax/x2' and keep rawId: 'x2', so the row keeps its identity. If it is a new model, retire x2 the usual way.
  2. Decide what reactor/xmax-x2-1 is, using Reactor's own docs, and add it to REACTOR_MODELS with its page slug.
  3. Remove the catch-all fallback in listModels. A fetch or parse failure, or a body with no models, should throw.
  4. Tests: a pricing body that lists xmax/x2 maps to the curated row and its rate. An unreadable body throws.
  1. tombeckenham assignedtombeckenham

  2. tombeckenham linked this to #689

  3. Fixed by #689, deployed to prod as 21f71a38 at 2026-11-17 22:33 UTC.

  4. tombeckenham closed this

Sign in to join this conversation.