1. Scene hook

Last month I was writing a brand proposal for a restaurant client, ran it through Claude for two days. Checked the bill at month-end — $30 gone. Heart was bleeding.

2. What it is + who's using it

Anthropic just dropped Claude Sonnet 5.5. Short version: their "mid-tier" got an upgrade — output quality now nearly matches the priciest flagship Opus, but at half the price.

I was on the old Sonnet 5 before — wrote proposals, edited copy, summarized meeting notes. Results were great, just expensive. Then over dinner with my freelance translator friend Xiaolin, she told me she uses flagship Opus for long contracts but runs daily volume on the old Sonnet. "It's not that I can't afford the expensive one — I just don't need to eat it every day."

Anthropic's own benchmarks: 5.5 is 30% faster than the last gen, big jump on proposals and knowledge work, some tests now match flagship Opus. I tested it for a week myself — same brand proposal, bill dropped from $30 to about $15. Writing copy, summarizing meeting notes — barely any difference. Highly complex technical proposals? Slightly weaker, but still usable.

This tool isn't for everyone — if your work is simple and your monthly Claude bill never crosses $7, don't bother. But if you're spending $15+/month, worth a look.

3. Replicate cost

  • Money: Charged by the tokens you send in. Single-task price halved — save 30-50% vs. before.
  • Time: 5 minutes to switch.
  • Tech barrier: Just use the Claude web app — no code needed.
  • First step: Open claude.ai → top-right avatar → model dropdown → pick "Sonnet 5.5".

4. Advice by stage

  • Just starting: Don't worry about saving this money yet — free credits will get you through the early phase. Get work flowing first.
  • Have 1-2 clients: Definitely worth switching. Half the price for 5 minutes of work — next month's bill will show it.
  • Scaling up: If AI costs you hundreds monthly, the savings cover a client dinner. What you save is pure margin.