Claude Code Subscription vs API Pricing: Opus 5.5 Guide

For one developer coding interactively most days, a subscription is usually cheaper and simpler. For CI pipelines, unattended agents, shared services or very uneven usage, the API gives clearer control and per-token accounting. This guide lays out current plan prices, API rates, how limits work, and a method for finding your break-even. Facts come from Claude's pricing page, Anthropic's API pricing documentation, the Max plan help article and Claude Code's model and fast-mode documentation, read between October 5 and October 30, 2026. Claude pricing, API pricing, Claude Code model configuration, Fast mode.

Subscription Plans That Include Claude Code

Every paid plan includes Claude Code, and your terminal sessions share the same pool as Claude on web, desktop and mobile. Claude usage limits by plan compares the session and weekly limits on Free, Pro, Max and Ultra. Limits reset on a rolling five-hour window, except on a verified Ultra account, and paid plans also have weekly limits. Anthropic does not publish a fixed message or token count, because usage depends on conversation length, the model you choose and the features you use.

API Rates for the Models Claude Code Uses

The Batch API halves these prices for asynchronous work. Five-minute cache writes cost $5 per million tokens on Opus 5.5 and $2.50 on Sonnet 5.5, and reading cached context is far cheaper than fresh input, which matters because Claude Code re-reads repository context constantly. Anthropic notes that Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text, so older token estimates understate current costs.

What Claude Code Runs by Default

From Claude Code v2.1.280, the default model is Opus 5.5 on Pro, Max, Team, Enterprise and the Anthropic API. Opus 5.5 and Sonnet 5.5 run at medium effort by default in Claude Code, and on the Anthropic API they use the 1M-token context window on every plan, including Pro, without a separate option. You can switch to Sonnet 5.5 or to Haiku 5.5, which the haiku alias selects on the Anthropic API from Claude Code v2.1.293, with /model to stretch your limits or lower API cost.

Fable models are never the default. Depending on your plan and seat, Fable usage can bill to usage credits rather than your included limits, and Claude Code asks for consent before a Fable request is billed that way in interactive sessions. On Max and Ultra, Fable is available within 50% of the weekly limits.

Fast Mode Is Always Extra

Fast mode is a research-preview configuration of Opus that prioritizes speed. On Opus 5.5 it costs $8 per million input tokens and $40 per million output tokens, twice the standard rate. On Pro, Max, Ultra, Team and Enterprise, fast mode is paid only from usage credits and never from your plan's included limits, even if you still have plan usage left. On Console accounts it is billed per token with other API usage and must be provisioned for your organization. Fast mode is not available on Amazon Bedrock, Google Cloud, Microsoft Foundry or Claude Platform on AWS.

When You Hit a Limit

On a subscription, reaching a limit gives you three choices: wait for the five-hour window or weekly limit to reset, move to a higher plan or seat, or turn on usage credits so you can keep working at standard API rates. Usage credits make a subscription behave partly like the API at the margin, so heavy months cost more than the sticker price. Claude's settings show where you stand at any time. Google Antigravity, which also offers Claude Opus 5.5 and Sonnet 5.5 on Google AI Pro and Ultra, limits usage the same way, with a five-hour quota under a weekly limit and paid credits beyond it (Antigravity limits by plan).

On the API, there is no session window; you pay for every token and are constrained only by rate limits and the spend limits you set. That makes cost directly proportional to work, which is easier to allocate across projects and teams.

Estimating Your Break-Even

To compare, estimate what a typical Claude Code task would cost on the API, then multiply by your monthly task count. The example below is illustrative, not an Anthropic figure: assume a substantial Opus 5.5 task reads 2 million tokens of context, 90% of it from cache, and writes 100,000 output tokens.

The per-task figure is 1.8 million cached tokens × $0.20 + 200,000 fresh tokens × $4 + 100,000 output tokens × $20, or $0.36 + $0.80 + $2.00 = $3.16. Sonnet 5.5 halves each rate, including its $0.10 cache read, so the same task costs $1.58. Your real numbers depend on repository size, cache hit rate and how much the agent writes, so measure a week of usage with the API's cost reporting before deciding.

Teams: Seats Versus Usage-Based Enterprise

For engineering teams, the choice shifts from personal plans to seats. Team Standard seats cost $20 a month billed annually and give more usage than Pro; Premium seats cost $100 and give five times Standard, and you can mix the two so only heavy Claude Code users get Premium. A ten-person team with three heavy users on Premium and seven on Standard pays 3 × $100 + 7 × $20 = $440 a month on annual billing, and gets central billing, SSO and admin controls.

Enterprise works differently: $20 per seat per month billed annually, plus all usage at API rates. That suits organizations whose usage varies widely between people, because light users cost little and heavy users pay for what they consume, with admin-set spend limits at user and organization level. Compare the two by estimating your team's total monthly token cost: if it is well below what Premium seats would cost, Enterprise can be cheaper; if most developers are heavy users, fixed Team seats are more predictable. Claude Team vs Enterprise compares the two plans on seat price, usage and admin controls.

Reading Your Usage Before You Decide

Measure before you buy. On a subscription, Settings > Usage shows how close you are to your session and weekly limits and how much you have spent in usage credits this month. On the API, the Console's usage and cost pages break spend down by model and workspace, and Claude Code's /usage command shows your own consumption. A week of real data beats any estimate.

When a Subscription Wins

A subscription is the better deal when one person codes interactively, works within predictable hours, and would otherwise spend more than the plan price in tokens. It also bundles everything else in Claude, such as chat, Projects, Claude Design and Research, and requires no budget administration. Pro is the cheapest way to try Claude Code seriously; Max suits developers who hit Pro limits most days, and Ultra those who hit Max's five-hour or weekly limit.

When the API Wins

Use an API key when software runs Claude Code without a person present, such as CI checks, scheduled refactors or bots that serve several users; when you need per-project cost allocation; when usage is bursty, with heavy weeks followed by quiet ones; or when you need concurrency beyond a personal plan. The API also lets you use the Batch discount for work that can wait. For organizations, Enterprise combines seats at $20 a month with usage at API rates, which is effectively the API model with admin controls.

Mixing Both

Many developers use both: a Pro, Max or Ultra plan for interactive work, and a Console API key for automation. On Max, Ultra and Team, the plan's monthly API credits can pay for part of that automation when the key belongs to the linked Console organization. Keep the two budgets separate so a runaway automated job cannot drain the limits you rely on during the day, and so finance can see which spend is personal productivity and which is production infrastructure. Within a subscription, turning on usage credits with a monthly cap gives a safety valve without moving everything to the API.

Cost Controls That Work on Either Route

Choose the cheapest model that does the job: Sonnet 5.5 costs half of Opus 5.5, and Haiku 5.5, at $0.10/$0.50 per million tokens for prompts up to 100,000 tokens, suits subagents and quick tasks. Keep repository context stable so caching works. Use fast mode only when waiting costs more than double the token price. On the API, set organization and workspace spend limits; on subscriptions, cap usage credits. Review usage weekly at first, because agentic coding can scale cost faster than expected.

Final Check Before Choosing

Estimate your monthly tasks and their API cost using a week of measured usage. If that figure is below $20, Pro is likely enough; if it falls between $100 and $200, compare Max and Ultra with your tolerance for limits (Claude Max vs Ultra); above that, or for automation, use the API or an Enterprise plan with spend controls. For plan details, see our Claude Code pricing page and the guide to Claude Pro and Max vs API pricing.