What changed
Anthropic released Claude Opus 5.5 on September 22. Input now costs $4 per million tokens, down from $5. Output costs $20, down from $25. Cache reads — the part that adds up when AI reviews long documents or code — fell 60%, from $0.50 to $0.20 per million tokens. Anthropic says the model runs 30% faster than the old Opus 5 and costs 40% less on typical work. It also scores higher on most benchmarks than Claude Fable 5.1, the pricier model above it.
What you can do with it this week
Hand Claude the weekly report you normally write by copying five emails and three spreadsheets into a doc. Give it the source files, tell it the format, and let it draft the whole thing. The cheaper cache reads mean you can feed it the same project files every week without the bill climbing. Try it on: summarizing client threads, pulling data from invoices, drafting scope-of-work docs, or reviewing contracts. Pick a task that takes two hours and see if Claude cuts it to twenty minutes.
What to keep in your own hands
The final call on what you send to a client. The decision to approve a price or a deadline. Your customer list, your pricing strategy, and any data your industry regulator says must stay private. Set a spending cap in your Anthropic account before you run your first big job — Claude will stop when you hit the limit, not when the work is done. If a task depends on being right the first time, check the output yourself before it leaves your hands.