Anthropic Cuts Opus 5.5 Pricing by 20% and Adds Fast Mode Switch

Anthropic made Opus 5.5 cheaper — $4/$20 and a Fast Mode switch

$4 per million input / $20 per million output tokens. It was $5/$25, i.e. −20%
Cache read $0.20 (−60%), cache write $5 (−20%)

Anthropic promises up to ~40% savings on typical load compared to Opus 5. Fast/priority tier is twice as expensive — $8/$40, but up to 2.5× faster; the base model is “over 30% faster than Opus 5”

Practice

Artificial Analysis Intelligence Index — Opus 5.5 (max) = 58, first place. Sonnet 5.5 — 56, GPT-6 Astra — 53, Gemini 4 Argon — 53
From Anthropic: Terminal-Bench 4.0 — 66.4% vs 52.3% in Opus 5; CursorBench 4.0 — 57.8% vs 46.6%; GDPval-AA v2.1 — 1842 Elo; OSWorld 2.1 (computer) — 81.8%; Humanity's Last Exam — 67.7%

Thinking mode can no longer be turned off. The depth of reasoning is controlled only by the effort parameter: low / medium / high / max. Reasoning tokens are billed as output and reduce max_tokens, so “just answer without reasoning” will no longer work

Subscriptions

In Pro, Max, Team and Enterprise the five‑hour limits have been raised, plus they gave a consumable pass for speed, you can stock up and spend when needed

How to test

Fast Mode should be found as a toggle in the app/Claude Code, better enable it in interactive sessions rather than nightly pipelines, because you overpay 2× for speed

https://www.anthropic.com/claude-opus-5-5