Anthropic “credit balance is too low”: fix it, tell it apart from spend limits, and stop it recurring

The short version

• The Console organization that owns the API key has no usable prepaid credits. Buy credits for that organization; retrying won't fix it.
• In Claude Code on a paid plan, the usual cause is different: a stray ANTHROPIC_API_KEY is billing a Console organization instead of your subscription. Check /status before you buy anything.
• It arrives as 400 invalid_request_error, not the 402 billing_error you might expect, so status-based error handling treats it like a bad request.
• Auto-reload keeps you running but doesn't cap spend. Pair it with organization and workspace spend limits, and keep keys out of the Default Workspace, which can't have one.

1. The two usual causes

The organization is out of credits. Claude API, playground and Claude Code usage on a Console account is paid from prepaid credits. Anthropic's billing article: “If you run out of credits, you can no longer call the API or use the playground until you add more.” Balances belong to the organization, and every key and workspace in it draws from the same balance, so one busy project or a leaked key can empty it for everyone.

Claude Code is using an API key you didn't mean to use. Anthropic's Claude Code error reference lists both causes: the Console organization has run out of prepaid credits, “or Claude Code is sending your requests with a Console API key when you meant to use your subscription.” Its fix: run /status, check the API key row, unset ANTHROPIC_API_KEY in the shell and your shell profile, relaunch claude, and run /login if needed. Buying credits also makes the error go away, but then you pay API rates for work your plan already covers.

One more case: the monthly API credits included with Max, Ultra and Team plans cover the Claude API, Managed Agents, the Agent SDK and the playground, but not Claude Code. An organization holding only those credits still gets “Credit balance too low” in Claude Code.

2. Which error is it?

Anthropic has several ways of saying “no more requests”. They differ by status code, message and how access comes back (API errors, rate and spend limits):

Only the rate-limit row is worth retrying automatically. The official SDKs retry rate limits and 5xx errors twice by default; they don't retry a 400, but they will retry the spend-cap 429, which Anthropic says keeps failing until access resumes. Spend limits on the Claude Code workspace are checked separately and can come back as a 429 with retry-after instead of a 400.

3. Fix it now

1. Make sure you're looking at the organization that issued the failing key. Balances aren't shared across organizations.
2. Sign in to the Console with an Admin or Billing role and open Billing. Click “Buy credits”, enter an amount and confirm. Anthropic says purchased credits are available immediately.
3. Optionally set up auto-reload (next section).
4. Re-send the failed request. If it still fails after the balance shows credits, contact Anthropic support with the request_id from the error body.

Purchased credits expire one year after purchase, can't be extended and are non-refundable, so buy what you expect to use.

4. Auto-reload keeps you running; it doesn't cap spend

On the Billing page, under Credit balance, “Set up auto-reload” asks for When credit balance reaches (the trigger) and Bring credit balance back up to (the target). If your balance is already at or below the trigger, the button reads “Turn on and purchase” and you're charged right away. If you also have plan-included credits, those are spent first, and auto-reload watches only your purchased balance, so it can reload while included credits remain.

Auto-reload turns an outage into a card charge, which is the right trade for production, but it keeps buying for as long as something is spending. The controls that stop usage are spend limits:

• Organization spend limit: Billing → Spend limits → Adjust limit (or Set limit). It can't exceed your tier's monthly cap (Start $500, Build $1,000, Scale $200,000 a month; Custom tier has none).
• Workspace spend limits: give each project or agent its own workspace with its own limit, so one runaway doesn't drain the shared balance. Anthropic's Claude Code error reference recommends the same.
• The Default Workspace can't have a spend limit (workspaces docs). Keys left there are capped only by the organization limit. See Anthropic Default Workspace limits for moving them.

Spend limits are monthly: they reset on the 1st of each calendar month, while plan-included credits refresh on your billing cycle.

5. Handle it in your app

There's no machine-readable code for an empty balance today; the message text is the only signal (a feature request on the Python SDK asks for one). Match on the message, alert a human, and stop the job:

import anthropic

client = anthropic.Anthropic()
BILLING_400 = ("credit balance is too low", "reached your specified")

try:
    msg = client.messages.create(model=MODEL, max_tokens=1024, messages=messages)
except anthropic.BadRequestError as e:  # HTTP 400
    if any(s in str(e).lower() for s in BILLING_400):
        alert_on_call(f"Anthropic billing stop: {e}")  # your paging/Slack/email hook
    raise  # either way, the same request will fail again
except anthropic.RateLimitError as e:  # HTTP 429
    err = (e.body or {}).get("error", {}) if isinstance(e.body, dict) else {}
    if (err.get("details") or {}).get("error_code") == "enforced_spend_limit_reached":
        alert_on_call("Anthropic tier spend cap reached; paused until the 1st (UTC)")
    raise  # a normal rate limit: back off, honour retry-after, cap the retries

Two things this gets right: a billing stop pages someone within minutes instead of surfacing as user complaints, and agent loops don't keep hammering the API with requests that can't succeed. Message matching is fragile, so check it again when you upgrade the SDK.

Checklist for a production app on Anthropic prepaid credits

1. Auto-reload on, with the trigger set to at least a day or two of normal spend, and the payment method and Billing role owner known.
2. An organization spend limit set below your tier cap, at a number you'd be willing to pay in a bad month.
3. Each app or agent in its own workspace with its own spend limit; nothing running on Default Workspace keys.
4. Billing 400s and the spend-cap 429 handled apart from rate limits: no retries, an alert to a human.
5. Claude Code machines and CI checked with /status so a stray ANTHROPIC_API_KEY isn't billing the wrong organization.
6. Something outside the app that tells you the same day when spend is far above normal.

Want to check the rest of your settings? The free AI key limit audit covers Anthropic, OpenAI and OpenRouter in a few minutes (no sign-up, no key needed). The OpenAI equivalent of this error is covered in OpenAI credit_balance_exhausted; for agents that can loop, see Runaway AI agent cost.

Where KeyBark fits (and where it doesn't)

KeyBark reads your Anthropic organization's cost report (with an Admin API key) from our servers about every 15 minutes during the beta and emails you when a daily or monthly threshold you set is crossed, or when today's spend jumps above your 7-day average, Default Workspace spend included. If you opt in, it can set a named API key to inactive when a threshold is crossed. That covers the “spend is far above normal” item in the checklist.

Limits, honestly: KeyBark does not read your credit balance or auto-reload settings, so it can't warn you that credits are about to run out; your app's error handling and the Billing page do that. Anthropic's Admin API isn't available for individual accounts, so you need an organization in the Console. Checks run on a schedule, so an alert can arrive after money has been spent. KeyBark is not a proxy and can't block a request.

FAQ

Sources (read 1 Nov 2026): Claude Help Center How do I pay for my Claude API usage?; Claude API docs Errors, Rate limits (spend limits section), API credits for Max, Ultra, and Team plans and Workspaces; Claude Code docs Error reference. KeyBark is not affiliated with Anthropic.