OpenCode Zen: Jev ids keep firing pricing_lost because models.dev has no record for them #314
Send to my agentSymptom#
Since the first OpenCode Zen poll after #313 deployed (around 02:33 UTC on 2026-10-13), pricing_lost (parser_miss) fires on every opencode poll for jev-1.13 and jev-1.13-free. The stored cards are kept, as designed.
Cause (not a parser bug)#
- #313 moved Zen prices to models.dev (
https://models.dev/api.json,opencode.models.*.cost). models.dev has no record forjev-1.13orjev-1.13-free, so both rows getpricing: null.docs/source-silent/opencode.mdfrom #313 already notes these ids have no catalog record. - Both ids are still listed live on
https://opencode.ai/inference/v1/models. - The OpenCode console page https://opencode.ai/v2/docs/console/models/ (the replacement for
/docs/zen) still publishes both prices in its Pricing table (per 1M tokens):Jev 1.13= input $0.042, output Free;Jev 1.13 Free= Free / Free. That page serves HTML only (.mdgives 404). - The stored cards (source
https://opencode.ai/docs/zen, extracted 2026-10-08/09) match those prices exactly: $0.042/1M input with $0 output, and $0 for both. They are valid but no longer re-sourced. - Jev is served on
/inference/v1/systemone(a structured-decision API, not chat), soactivitystays null.
Decision needed#
For a listed Zen id that models.dev does not record but the console Pricing table prices, should the adapter:
- read the console page's Pricing table as a fallback source for ids models.dev lacks, keyed through the page's model-list table (display name to id), or
- state
absent: { pricing: 'cleared' }, as #311 did for Gemini ids the pricing page stopped naming? This would null two prices the provider still publishes, or - something else, such as waiting for models.dev to add Jev?
The ingest maintenance run did not change any code. Option 1 adds a second price source to a provider that #307 deliberately moved to one source. Option 2 drops a price the provider still publishes.
claude-opus-5-5-backup is also listed but missing from models.dev. It has no stored card, so it emits no signal.
Done when a full opencode poll cycle emits no pricing_lost for these ids and the stored cards follow the chosen rule.
Decision (2026-10-18)#
Read OpenCode Zen prices from the console models page (services/www/src/docs/content/console/models.mdx in anomalyco/opencode, the "Pricing" table, per 1M tokens) as the sole price source, replacing models.dev cost for Zen. Fail closed: a page that parses nothing throws rather than nulling stored cards. Ids the table does not name get absent.pricing: 'cleared' (as #315). Go prices stay null.
tombeckenham closed this
Sign in to join this conversation.
Production check after #315 deployed: the last
pricing_lost(parser_miss) forjev-1.13/jev-1.13-freefired at 05:18 UTC on 2026-10-13. OpenCode polls kept running after that, the latest at 06:48 UTC, and none fired the event again./v1/models/opencode/jev-1.13now servespricing: null, as #315 intended.This issue is still open for one question that needs a person: the v2 console models page still publishes prices for these ids. Should the adapter read that page instead of clearing the price? That is a product decision, so this automated check leaves the issue open.
Created with PostHog Desktop