Claude Code Pricing: Opus 5, Plans, Credits, and API Rates

Buying path

For human-steered coding, begin with an eligible Claude subscription and sign Claude Code in with the same Claude account. Pro is the lowest paid individual entry, while Max buys a larger included capacity envelope. The Free Claude plan is useful for evaluating the broader assistant, but it does not list Claude Code as included access.

Treat authentication as part of the purchase decision. A subscription session draws from limits shared with Claude, whereas a Console credential or an ANTHROPIC_API_KEY moves work to separately metered billing. Check the active account and model inside Claude Code before a paid workload; the terminal interface alone does not reveal which budget owns the request.

Opus 5 access should also be verified in the actual model picker. Anthropic makes it the default for Max, Team Premium, usage-based Enterprise, and Anthropic API accounts, while Pro, Team Standard, and subscription-seat Enterprise default to Sonnet 5. Organization settings, provider routing, and explicit model choices can change that result, so buying a plan does not by itself pin every session to Opus 5.

Upgrade triggers

Move to a higher individual plan or a Premium team seat when real work repeatedly reaches the included limit and waiting for a reset interrupts delivery. Claude and Claude Code share subscription capacity, and consumption changes with repository context, conversation length, model, effort, tools, and parallel sessions. A Max multiplier or Premium allowance is not a package of API tokens and should not be converted into one.

Usage credits are the cleaner bridge when overflow is occasional. Standard-mode work after an included limit is charged separately at standard API rates, while the normal plan window continues to reset on its own schedule. Set a conservative spend cap before enabling credits and compare several representative sessions with the cost of a higher fixed tier.

Opus 5 defaults to high effort in Claude Code and supports low, medium, high, xhigh, and max. Higher effort can improve demanding work while consuming more tokens; max is normally session-only. Extended context is another separate trigger: Opus with 1M context is included on Max, Team, and Enterprise, but Pro uses paid credits for that route. Do not infer the Messages API’s 128K output ceiling as a guaranteed output size in the Claude app or every Claude Code surface.

Fast mode is a latency purchase, not a capacity upgrade. It uses Opus 5 or Opus 4.8 at premium token rates from the first Fast-mode token, even when subscription allowance remains. Use it only where faster feedback has measurable value; standard mode remains the safer default for long autonomous work, CI, or background execution.

API and team boundaries

Claude subscriptions and Claude Console are separate commercial systems. Console-authenticated Claude Code meters usage from the first request, and standard Opus 5 rates are based on input and output tokens. That route fits unattended automation, shared budgets, and workloads that need per-workspace spend controls; a paid app plan neither funds nor grants general API access.

The API contract also needs its own migration review. Opus 5 uses a 1M context window and supports up to 128K synchronous output through the Messages API, but adaptive thinking is on by default and max_tokens covers thinking plus response text. Disabling thinking is accepted only through high effort; combining disabled thinking with xhigh or max returns an error. Web fetch and Priority Tier are not supported for Opus 5, so do not carry an older model configuration forward without a fresh effort, output-budget, and tool-availability test.

Team seats place Claude Code inside a managed workspace and share each seat’s allowance across Claude surfaces. Current Standard and Premium seats both include Claude Code, with Premium carrying more usage; owners control paid overflow. Current usage-based Enterprise instead combines a seat charge with consumption at API rates and has no per-seat included limit, while older seat-based contracts can follow different rules. Confirm the contract type before comparing Enterprise with Team.

Programmatic use has one more boundary. Anthropic paused its announced separate Agent SDK monthly credit, so Agent SDK, claude -p, and third-party Agent SDK app usage still draw from subscription limits for now. Production automation should therefore have an explicit identity and budget rather than relying on a proposed credit or an interactive user’s allowance.

Final pricing check

Before paying, verify the billing interval, active login, default model, allowed-model policy, effort level, context entitlement, and who can authorize overage. Annual prices can be displayed as monthly equivalents while the full commitment is billed upfront, and Team or Enterprise spend can combine seat and usage ownership across different administrators.

For standard metered work, confirm the model ID, input and output rates, caching behavior, workspace limits, and migration settings. For Fast mode, confirm that the session is on a supported first-party route, that usage credits or Console billing are enabled, and that the interface supports the mode; the Visual Studio Code extension and listed partner cloud routes do not.

Finish with a small representative pilot under the exact account and credential that will pay. Measure subscription limit pressure separately from token-billed spend, inspect the active model and effort, and review the import diff before publication. The safe route is the one whose identity, allowance, overage, and output boundary are explicit before the coding workload scales.