Anthropic released Claude Opus 5.5, its flagship reasoning model, on September 22, 2026, cutting API prices 20% and cache-read costs 60% at once. The move lands days after OpenAI halved its own frontier model prices, hitting the model tier many AI support agents reach for when a reply needs real reasoning, not a canned answer.
What changed
Per Anthropic’s pricing documentation, Opus 5.5 now costs $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5 — a 20% cut both ways. Cache writes drop too: $5 per million tokens for a 5-minute cache (from $6.25) and $8 for a 1-hour cache (from $10). Cache reads fall hardest, 60%, from $0.50 to $0.20 per million tokens. Anthropic says the model “performs at the level of Claude Fable 5.1 on most work” while costing “40% less to run than Opus 5.” A research-preview fast mode runs up to 2.5x faster at $8/$40 per million tokens, API-only. Opus 5.5 is live now as claude-opus-5-5 on the Claude API, AWS, Google Cloud and Microsoft Azure.
Why it matters for Thai sellers and support teams
Opus is Anthropic’s top reasoning tier, not what most FAQ bots run on — it’s the tier a business reaches for when AI needs to read a full order history, judge a refund case, or decide when to hand off to a human before replying. A 20–60% cut there makes it more affordable to let AI take on harder conversations, not just easy ones. It also follows OpenAI’s own 50% API price cut days earlier — capable AI support is getting cheaper across every major vendor at once.
What to do now
- If a vendor bills per AI reply rather than a flat fee, ask whether this cut has reached your invoice.
- Testing AI for complex escalations rather than simple FAQs? Opus 5.5 is worth a fresh cost comparison.
- Keep high-volume, simple replies — order status, opening hours — on a cheaper model tier; this cut doesn’t change that math.
- If a developer switches your integration, the model ID to look for is
claude-opus-5-5.
