T
@theo
Planned!
62
3
2.6K
@theoposted on X
It is genuinely insane that Anthropic will bill you differently if you mention certain words in your prompt or have certain files in your codebase
View original tweet on X →
Community Sentiment Analysis
Real-time analysis of public opinion and engagement
Sentiment Distribution
80% Engaged
73% Positive
Positive
73%
Negative
7%
Neutral
20%
Key Takeaways
What the community is saying — both sides
Supporting
1
Opaque, hidden billing triggers
users report being charged extra by silent string-match heuristics (examples: HERMES.md, THEO.md) and even auto-switched to API rates while on subscription, creating unpredictable bills.
2
Billing can be weaponized
several replies warn this is effectively a prompt-injection-for-your-wallet — a compromised skill or repo name could be used as an attack vector to inflate costs.
3
Trust in Claude/Claude Code is eroding
developers say unexpected heuristics cut pipelines mid-run, making code assistants unreliable — “you shouldn’t have to worry about a commit string triggering API rates.”
4
Safety claims look hypocritical
people call out Anthropic’s “Safe AI” messaging while accusing them of exploitative, heavy-handed monetization and poor support.
5
Users will (and already can) defect
multiple replies advocate moving to open models or competitors (Codex, GPT 5.
6
because opaque charging and aggressive filtering push builders away.
because opaque charging and aggressive filtering push builders away.
7
Fixes are straightforward and demanded
practical suggestions include showing in the dashboard which heuristic flipped before billing, clearer pricing rules, and explicit opt-ins for auto-overage charging.
8
It’s a slippery slope
people fear pricing tied to “workflow shape” (file names, variable names, use cases) will expand — “soon they’ll ban certain variable names or charge by use case.”
9
Customer support failures amplify anger
reports of no warnings, no appeals, withheld refunds, and post-payment identity hurdles make the issue feel predatory, not accidental.
10
Some react with sarcasm or opportunism
replies range from calls to prank orgs and “regexmaxxing” for free sequences to gloating about exposing the system’s quirks.
Opposing
1
unexpected $20 charge
from X/Grok despite having no subscription.
2
unreviewed third‑party “vibe coded” apps
, warning that buggy code — especially in apps built for government or defense — can harm end users.
3
Hermes
are “cooked” (joke about being affected).
4
not reproducible
for them; they note extra usage can sit unused and question how billing works for users with no wallet balance (claim: extra usage cannot be auto‑deducted).
5
context routing
, not keyword billing — larger context windows hit higher‑priced inference tiers, which can raise charges.
Top Reactions
Most popular replies, ranked by engagement
E
@eddiboi
need a video just on this
19
1
2.7K
W
@WillToMake
Wait til they take your $200 and THEN ask for 2 forms of government ID or you can't use the services you already paid for @bcherny @_catwu how do I get my $200 back? Moved exclusively to Codex
15
0
660
P
@PhiloGroves
People who are actually named Hermes are cooked
2
0
514
A
@aias_0
It’s context routing, not keyword billing. Bigger windows just hit pricier inference tiers.
2
0
122
S
@somaco_sf
I was randomly billed $20 by X/Grok yesterday even though I dont have a subscription plan...
1
0
754
Report an Issue
Found something wrong with this article? Let us know and we'll look into it.