Hello forum, i’ve noticed some strange usage with Claude 4 Sonnet—the price is INSANE, yet i only worked on 10k lines of code.
Has anyone else noticed any unusual pricing? is that normal?
1 Like
Yeah, I’ve noticed the same thing — I was also getting charged absurd amounts in relation to token usage. 
One thing that helped me reduce costs while still getting good results was being very intentional with how I structure prompts. For example, I try to reference relative paths, mention specific variable names, and prepare a detailed context prompt that clearly explains what the model needs to do.
By doing that, I usually manage to stay under $1 per interaction — but even with those precautions, I still think the pricing is kind of insane right now. 
1 Like
Hey @Lorenzo_Tavola, my understanding is that this is a UI discrepancy.
You aren’t actually charged for these calls. I won’t share your billing specifics here in public, but if you check your billing from June, you’ll see:
token-based usage calls to non-max-claude-4-sonnet-thinking
The cost for each request should be around $0.04 for June, which should have been finalized on July 1st. Your spend limit will have also reset on July 1st.
If this is the case and I wasn’t changed what it shows, thats much better..
IF I look at billing and take it at its word, then it does indeed reflect this.
I want back to usage for June to double check this and I noticed a few things.
- The amount of rows it says its showing is incorrect (and inconsistent)
- 100 rows per page shows 46 on page 1 and 41 on page 2, you get the idea.
- 500 rows only shows 2002 on page 1 and 194 on page 2.
- If I could what I have in my usage vs billing, they are 2 different numbers - and used is including whats supposed to be included in Pro.
Are updates to the website even being tested? It has had so many bugs lately, whats happening here?
Thanks for all the feedback—here’s an update with the full numbers for June and July.
June 2025
- 1,054 claude-4-sonnet-thinking requests @ $0.03 each → $31.62
- 87 extra-fast requests @ $0.04 each → $3.48
- 2 token-based GPT-4.1 calls → $0.52
- “Token-based usage to non-max claude-4-sonnet-thinking” → $14.99 (looks like the monthly plan)
- Mid-month usage already paid → -$20.00
Final total: $50.61 (of which $30.61 is still outstanding)
July 2025 (as of 2 July)
- 7 sonnet-thinking requests → $0.98 (that’s $0.14 each—why?)
- “Token-based usage to non-max claude-4-sonnet-thinking” → $14.85
Running total: $15.83
Issues I’m seeing
- The live “Upcoming invoice” widget shows wildly inflated numbers during the month (other users confirm this).
- The same activity is billed twice—once as a request and once as token-based.
- Usage pagination is broken (missing rows / inconsistent counts).
- Even after the final adjustment I’m still paying $50 for a weekend’s work on ~10000 lines of code—unacceptable.
Is anyone else seeing similar numbers (maybe with Opus), or am I the only one getting hammered like this?
I notice while Claude 4 is “thinking” on a single question, the price ticker keeps jumping. In other words, one prompt can spawn several internal “thinking” steps, and each step is billed as a separate task. That means the total cost of a single request can climb in real time while the model does its own chain-of-thought.
This makes the final price completely unpredictable—and explains why even a quick Q&A can end up far above the advertised $0.03 / request.
3 Likes
I thought it was a bug because I was under Student Pro Plan, which is supposed to be free. However, even before Pro Plan’s limit has reached, the spending limit has been filled. This happened without any precaution and quietly like a ninja.
That’s it? I want to see what caused this so badly.
Weird thing for me is that it started charging even before Pro Plan’s limit has reached.
I checked to see why and it just says:
“token-based usage calls to non-max-claude-4-sonnet-thinking, totalling: $52.28”
1 Like