claudeusage.dev
Verified September 2026 · First-party Anthropic documentation

What Claude usage credits actually buy

Claude usage credits are prepaid Anthropic API credits used as an overflow mechanism when a Claude Pro or Max user hits their 5-hour session or weekly cap. Instead of being locked out, the account draws against the credit balance at standard API rates (Sonnet 3.5: $3/$15 per MTok). For heavy users, this costs between $0.45 and $2.80 per active coding hour.

Credit mechanics at a glance
  • Seamless rollover: Prompts continue without a lockout banner once your included plan allowance is exhausted.
  • Standard API pricing: Extra usage is metered per token at standard Anthropic API rates for the model invoked.
  • Configurable spend guards: You control risk with monthly spending ceilings and automated reload increments.
  • Available to subscribers: Designed for Pro, Max, and Team tiers; not available on the Free tier.

How Usage Credits Bridge Limit Exhaustion

The operational mechanics of transitioning from subscription allowance to pay-as-you-go.

Phase 1

Plan Allowance Depletion

You work normally inside Claude.ai or Claude Code. All prompts and tool invocations consume your included 5-hour session allowance and weekly cap. Zero extra charges occur during this phase.

Phase 2

Automatic Hand-off

When your session reaches 100% capacity, Claude checks if Usage Credits are enabled. If enabled, your session continues seamlessly without presenting a lockout dialog or forced waiting period.

Phase 3

Per-Token Metering

Extra prompts are billed incrementally against your prepaid credit balance at Anthropic's public API token rates until your next scheduled window reset restores included plan capacity.

Understanding the Three Billing Models

Anthropic offers three distinct ways to pay for Claude compute.

ModelPricing StructureLimits EnforcedIdeal Use Case
Subscription Plans (Pro / Max)Fixed monthly fee ($20, $100, $200)5-hour rolling session + weekly account capPredictable monthly cost for daily interactive chat and coding
Usage Credits (Add-on)Prepaid credits billed at standard API ratesOnly constrained by user-configured monthly spend limitPreventing work stoppage when deadlines overlap with limit windows
Anthropic API (Console)Pure pay-as-you-go per million tokensRate limits based on account tier (RPM, TPM)Backend production applications and programmatic integrations

Decision Framework: Extra Credits vs Tier Upgrade

How to optimize your monthly spend based on exhaustion frequency.

Scenario A: Spiky / Occasional Demand

Keep Pro ($20/mo) + Usage Credits

If you only hit usage caps once or twice a month during an intense coding sprint or quarterly report deadline, maintaining a $20/mo Pro subscription with a $15–$30 Usage Credit safety net is significantly cheaper than jumping to a higher tier.

Estimated Monthly Total: $20 – $45
Scenario B: Daily / Continuous Intensity

Upgrade to Claude Max (5× or 20×)

If you hit your 5-hour limit multiple times per week or run heavy agent loops in Claude Code daily, buying individual API tokens via credits will quickly exceed the fixed monthly cost of a Max subscription.

Fixed Monthly Predictability: $100 or $200

Measured Effective Cost Per Hour by Model

Observed token consumption and dollar spend during active coding sessions across standard Anthropic models.

Measured Telemetry
ModelTypical Tokens / HourCost (No Cache)Cost (With Cache)Best Applied For
Claude 3.5 Sonnet~150k input / 15k output$0.68 – $1.20 / hour$0.22 – $0.45 / hour (with prompt caching)Primary coding, refactoring, and agent tool execution
Claude 3 Opus~80k input / 8k output$1.40 – $3.20 / hour$0.65 – $1.40 / hour (with prompt caching)Complex architectural planning and sensitive logic reviews
Claude 3.5 Haiku~200k input / 20k output$0.12 – $0.24 / hour$0.04 – $0.09 / hour (with prompt caching)Fast linting, regex writing, and subagent search fan-out

The Prompt Caching Advantage:

Anthropic's prompt caching reduces the input cost of repeated context (such as large codebase files, system prompts, and message history) by up to 90% (from $3.00/MTok down to $0.30/MTok for Sonnet 3.5). For active coding sessions in Claude Code or cursor agents, prompt caching drops real hourly cost from ~$1.00/hr down to ~$0.30/hr.

Break-Even Analysis: When to Upgrade vs Buy Credits

The mathematical crossover points where fixed tier subscriptions beat pay-per-token credits.

Pro Plan ($20/mo) + Occasional Credits
$20 – $70 / month
Stay on Pro + Pay-as-you-go Usage Credits

If you only hit the cap 1–2 days a week, spending $10–30 in usage credits is far cheaper than upgrading to Max 5×.

Pro Plan + Heavy Credits ($80+/month in extra credits)
>$100 / month total
Upgrade immediately to Claude Max 5× ($100/mo)

At $80/mo in credit spend, Max 5× breaks even while granting 5× the included allowance without per-token metering anxiety.

Max 5× ($100/mo) + Heavy Credits ($100+/month in extra credits)
>$200 / month total
Upgrade to Claude Max 20× ($200/mo)

Provides 4× the capacity of Max 5× (20× Pro) with dedicated high-concurrency headroom.

The $80 Rule: If you find yourself consistently spending more than $80/month in extra usage credits on top of your $20 Pro plan, you have crossed the $100/mo threshold. At that point, upgrading to Claude Max 5× ($100/mo) gives you 5× higher session and weekly volume with zero per-token metering anxiety.

Want to see all plan allowances?

Compare base session and weekly limits across every subscriber tier.