An email sent me scrambling to check my bill

I was scrolling my feed this morning when I saw OpenAI and Anthropic releasing new models within 90 minutes of each other — and both cutting prices. First thought: "Marketing gimmick, right?" Second thought: "Wait, how much money is left in my account?"

I've been tracking every monthly AI tool expense in a spreadsheet since last year — GPT for content generation, Claude for copy polishing. Peak month hit over 600 RMB. Watching everyone else drop prices, I was sweating too.

What actually happened

Short version: The two strongest AI companies released upgraded versions on the same day, and cut per-unit prices in half.

Anthropic released Claude Opus 5.5, which topped the industry benchmarks and costs 40% less than the previous generation. OpenAI fired back with GPT-6 in Sol and Luna versions, claiming "half the price, no shrinkage." For us non-coders, the most direct meaning is — doing the same task next month might cost you half as much.

My friend Xiaocheng (he runs a Xiaohongshu e-commerce shop) uses GPT to write 50 product descriptions daily. Last month he spent 380 RMB on GPT alone. Same workload next month might cost him 190. Yesterday in our group chat he dropped a "Going to renew for three months and try it, you're on your own" — with a screenshot timestamped 1 AM. This is real. I've seen him comparing AI tools on his phone in the subway.

What I can do today

Replication cost, broken down:

  • Money: 0 RMB to start. Switching models on an existing account costs nothing.
  • Time: 5–10 minutes to read the new version notes.
  • Technical barrier: No coding needed. The web chat box is enough to feel the difference.
  • First step: Log in at chat.openai.com or claude.ai, look in the model dropdown for anything marked "new" or "Sol/Luna/5.5".

My own first step: Take the product copy I had GPT rewrite last week, run it through the new version, compare quality, then decide whether to switch. I got lazy on this before and paid for it — switched everything over without testing and found the new model was actually more "oily" in one specific scenario.

How to play this at each stage

If you're just starting out, still on the free tier: No rush to test now. Free credits aren't moving. Come back to this when you finally pay.

If you already have 1–2 clients and spend 100–200 RMB a month: I'd spend 10 minutes today running a side-by-side — same task, old model vs new model (think "write a Xiaohongshu caption"), compare quality, then decide. If the new one holds up, switch and pour the savings into ads.

If you're scaling and about to plug into automation for batch runs: Half-price means double the tasks on the same budget. I'd build a small flow and run it for a week first, watching whether error rates climb (price drops sometimes come with rate limits). Only switch everything over after that checks out. And don't waste just because it's cheap — I made this mistake too. Bill went down but token usage doubled, so I actually spent more.