The bug that cost $200
This week, a GitHub issue titled HERMES.md went viral on Hacker News: an Anthropic billing bug silently charged a developer $200 extra — and Anthropic initially refused a refund.
The thread now has 289 comments and 737 upvotes. Most of them are developers sharing their own billing anxiety stories.
Here's a sample of what people are saying:
"I disabled my Claude Code usage because I was terrified of runaway charges."
"Metered billing for AI tools is the new dark pattern."
"I now check my Anthropic dashboard more than I check my bank account."
This is not a bug story. This is a structural problem with how AI is billed.
The metered billing trap
Every major AI provider — OpenAI, Anthropic, Google Gemini — bills by the token. This means:
- A runaway agent loop can cost you hundreds overnight
- A misunderstood context window multiplies your bill silently
- A billing bug, like the HERMES.md case, is nearly impossible to audit after the fact
- You check your API dashboard like a gas pump meter, watching the numbers tick up
This is fine for enterprises with finance teams and spend alerts. It is brutal for individual developers, students, and freelancers.
What flat-rate billing actually looks like
Here's the alternative model:
# Call Claude API — SimplyLouie flat-rate
curl https://api.simplylouie.com/v1/messages \
-H "x-api-key: YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-4-5",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Explain this code to me"}]
}'
You pay $2/month. That's it. No token counters. No usage dashboards. No surprise charges. No billing bugs that require a GitHub issue and 289 comments to resolve.
The response is identical to calling Anthropic directly — same Claude model, same response format, same API structure.
Why this matters for developers outside the US
The HERMES.md developer got a $200 charge. For a developer in:
- Nigeria: that's 2-3 weeks of salary
- India: Rs 16,500 — nearly a month's rent in many cities
- Philippines: ₱11,200 — over a month's worth of groceries
- Indonesia: Rp3.2M — a significant emergency fund hit
Metered billing doesn't just cause anxiety. For developers in emerging markets, a billing bug like HERMES.md is financially catastrophic.
Flat-rate billing at $2/month is not a price feature. It's a safety feature.
The technical argument for flat-rate
Some developers argue that metered billing is more "fair" because heavy users pay more. This is true at scale. But it creates a hidden cost that never shows up in the pricing page: cognitive overhead.
Every time you use a metered AI API, you're making micro-decisions:
- Should I include the full file in context, or trim it?
- Is this prompt worth the tokens?
- Should I stream this response or batch it?
- Did that agent loop cost me $0.20 or $2.00?
That cognitive overhead is a tax on your productivity. Flat-rate billing removes it entirely. You just... use the API.
What to do if you got an unexpected Anthropic charge
- Go to console.anthropic.com/settings/billing
- Export your usage log for the disputed period
- Open a support ticket with the exact timestamp of the anomaly
- Reference the HERMES.md GitHub issue — Anthropic is now aware and has acknowledged the bug
- If refused, dispute via your credit card issuer (this works, metered billing disputes are winnable)
The alternative
If you want Claude API access without metered billing anxiety:
SimplyLouie — $2/month flat rate, Claude API, no usage meters, 7-day free trial.
For developers who can't afford to check their billing dashboard with their morning coffee — this is what we built.
The HERMES.md issue is at github.com/anthropics/claude-code/issues/53262. The Hacker News thread is worth reading — it's one of the most honest conversations about AI billing transparency in months.
Top comments (0)