[BUG] Billing between claude.ai and claud code is confusing.

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

API key works — account just has no credits. Go to console.anthropic.com → Plans & Billing and add some.

For 750 files with Haiku at ~500 tokens each, expect roughly $0.10–0.20 total. Adding $5 will cover this run many times over.

I have 469.51 balance

The key and the credits are in different workspaces. This happens when you have multiple organizations in console.anthropic.com.

Check: console.anthropic.com → (org switcher, top left) → make sure the workspace with $469 is selected → API Keys → create a new key there.

Replace the key in your .env with the new one from the correct workspace.

How do I know which workspace this is in?

In console.anthropic.com, the org/workspace switcher is in the top-left corner — it shows the current workspace name next to a small dropdown arrow.

Your $469 balance will be visible under Plans & Billing only in the workspace that holds it. Switch workspaces until you see the balance there, then create the API key from that same workspace.

If you only see one workspace and the balance shows $469 but the API still returns "too low", it may be a billing tier issue — some accounts require a minimum purchase to unlock API access even with credits. In that case, contact Anthropic support.

There is a Default and Claude Code workspace. There is no Plans & Billing in either.

You're on a Claude Code subscription plan — billing is handled differently, no credit top-up UI. The $469 balance is likely in a separate personal/legacy account.

Try this: switch to the "Claude Code" workspace (top-left dropdown), then go to Manage → API Keys and create a key there. That workspace is tied to your active subscription and most likely has API access.

Replace the key in .env with that one and re-run the fixture test.

Can't create key. says
Create an API key
To create an API key in the Claude Code workspace, install Claude Code and then login with your Console account
Create a key to integrate with the Claude API. You can use the API directly or through a client SDK.

The Claude Code workspace is locked for API keys — it's only for Claude Code itself.

You need credits in the Default workspace. The $469 you saw is likely in a different account (work account, or a separate login).

Easiest path: add $5 to the Default workspace.

Switch to Default workspace in console.anthropic.com
Look for Settings or a billing link — it may not be in the sidebar but accessible at console.anthropic.com/settings/billing directly
If you can't find billing there either, the $5 option is to go to console.anthropic.com/billing while in the Default workspace — Anthropic recently moved billing out of the sidebar in some account types.

No...I have f'ing $400+. I want to know how to use that, not spend more.

Fair enough. Where did you see the $469 balance — was it on claude.ai, the console, or somewhere else? That'll tell us which account it's actually in.

That's claude.ai extra usage credits — for the chat interface. It's a completely separate billing system from the API. They can't be used for direct API calls from scripts.

Your options:

Add $5 API credit to console.anthropic.com (Default workspace → billing). The classifier will cost ~$0.15 to run. Annoying that it's separate, but it's a one-time setup.
Restructure the classifier to run inside Claude Code — instead of direct API calls, it generates a prompt you paste here, or we build it as a slash command. More work to build, but uses your existing subscription.
Use a free local model (Ollama + llama3) for classification — no API costs ever, but lower accuracy on edge cases.
Honestly, option 1 is the right call for $5 unless you're philosophically opposed to it. Option 2 is interesting but overkill for a one-shot cleanup job.

Fucking A. 2

Simplest version of option 2: I do it right here, right now. I read all the files, classify them in batches, and write manifest.tsv directly. No API key, no architecture changes, uses this session.

750 files × ~200 chars each is manageable in batches. The apply.py stays exactly as-is.

Want me to proceed? It'll take a few minutes and a lot of tool calls.

It isn't a one shot job. I have a lot of other projects/work to do through Claude Code

Right — Option C then: rewrite classify.py to call claude --print via subprocess instead of the Anthropic SDK. Same taxonomy, same manifest output, same apply.py. Just replaces the API calls with batched calls to the Claude Code CLI you already have.

Batches of 20 files per call → ~38 subprocess calls for 750 files. Reusable any time.

Read a file, created a file
That sucks.

Edited a file, ran a command
Yeah, it does. The Anthropic billing split between claude.ai and the API is genuinely annoying — your $469 is trapped in one system, unusable in the other.

The claude --print approach works around it but has a tradeoff: each batch spawns a full Claude Code subprocess, so 750 files at 20/batch = ~38 sequential calls. Probably takes 5-10 minutes vs 30 seconds with the API. For a one-time cleanup it's fine; for something you run weekly it'd be irritating.

Test it now:

uv run classify --vault fixtures --source . --out fixtures/test-manifest.tsv
The reason I have so much money there is because that is where you interface lead me anytime I ran out of tokens. The system should have had me add value to where the money was needed

That's a legitimate complaint — the UI upsells extra usage on claude.ai without making it clear that it's completely siloed from API access. You paid into the wrong bucket because the product never told you there were two buckets.

Worth filing at https://github.com/anthropics/claude-code/issues or the feedback link in claude.ai — that's a real UX failure and Anthropic should hear it.

What Should Happen?

Anthropic should make sure its not confusing to know what I am adding money to.

Error Messages/Logs

Steps to Reproduce

I don't know how I got to this point where I was working, I get told to add money and then it doesn't work in the environment I'm in.

Claude Model

Not sure / Multiple models

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

Version 1.3109.0 (35cbf6) Claude for WIndows

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_