Haiku 5.5 at max reasoning burns more tokens than Opus 5.5 max, user finds
haider1 · x · 2026-10-08
User haider1 flags a cost quirk with Anthropic's Haiku 5.5: cheap per task but extremely token-hungry — at max reasoning effort it consumes more tokens than Opus 5.5 max, over 2x more than fable 5.1 max, and far more than the astra models.
He asks where this tradeoff actually makes sense, inviting real-world use cases.
More from Models
- How Long Before Anthropic Ships a Decision Model Like Jev? — tristanbob · 2026-10-08
- Reddit user spots unconfirmed Gemini 4.1 appearing in the app, slow for a Flash model — iamsultanakbar · 2026-10-08
- Jev hits 32% of OpenRouter classification requests in 2 weeks; orgs now juggle 12 models — Tiancaixinxin · 2026-10-08
- Liquid AI open-sources d1 decision models: 3B model answers in one forward pass, 8ms on RTX 4090 — huggingface · 2026-10-08
- Hands-on: OpenAI's new Decisions API edges out Jev in accuracy and image classification — cantrell · 2026-10-08
- New Haiku matches last year's frontier, meaning Claude Code-class capability is heading toward 'too cheap to meter' — johnvmcdonnell · 2026-10-08