What happens#
Since 23:06 UTC on 2026-11-19, the Perplexity models poll has logged ingest_failed with the error perplexity: effort level names no model (job models-poll).
The effort page (https://docs.perplexity.ai/docs/agent-api/effort.md) now states the xhigh level like this:
> [accordion] xhigh — current effort level values
> * **Model:** `openai/gpt-6.2-sol` (no fallback model)
parsePerplexityEffortLevels (src/server/providers/adapters/perplexity.ts:271-291) reads a Model: bullet only when it is a bare backticked id (/^`([^`\s]+)`$/). The trailing (no fallback model) therefore leaves model undefined, and the parse throws. The other levels are unchanged:
fast uses Models: with a fallback.
low, medium and high each use a bare Model:.
ultra still says it `has no frozen configuration` and now names anthropic/claude-opus-5-5 in its prose.
Fix#
- Accept
`id` (no fallback model) as a Model: bullet that names id. That is a stated fact (there is no fallback), not an unread shape.
- Any other text after the id should still throw.
ultra stays model-less for flags: no frozen configuration means no stated settings to bind. Naming its runtime model in prose does not change that.
- Add a fixture test from the live page.