Guides

    Claude Pricing 2026: What Claude, Claude Code and the API Cost for Real Tasks

    What Claude, Claude Code and the API cost for real tasks, with the cheapest sensible plan per profile. Prices checked on 09-11-2026.

    November 9, 202612 min read
    Share:

    Published 09-11-2026. Last verified on 09-11-2026 and reviewed monthly.

    Quick answer

    This page is for the person who has to choose a Claude plan and predict the monthly bill before buying: a solo developer, the lead of a small dev team, a marketer on a Pro plan or an engineer who will put Claude behind an API. Our verdict in one line per profile. A solo developer starts on Pro at $17 a month billed annually ($20 monthly) and moves to Max at $100 when the session limit gets in the way. A five-person dev team costs $100 to $125 a month on Team Standard seats. A marketer who runs a weekly routine stays on Pro. An API builder pays per token, and the model choice moves that bill more than anything else: the same ticket classifier costs $1.10 a month on Haiku 5.5 and $44 on Opus 5.5. All prices below are in USD and were read on 09-11-2026.

    How we checked these numbers

    Our gate for this page: a figure goes in only if we opened the vendor page and dated it, and a task cost goes in only if every rate in it comes from such a page and the result is a dollar figure for one named job. We label each number as vendor-stated, calculated by us or an assumption. We did not run these tasks ourselves, so every token count below is an assumption and every cost is calculated by us from list prices. We say "not tested by us" wherever that is the honest label.

    Three things failed our gate and stay out. We do not quote a price for a Max 20x tier, because the pricing page no longer lists one: it shows Max at $100 and Ultra at $200. We make no claim that Cowork is included in Pro or Max, because the page does not state it. We give no figure for Enterprise bills above the seat minimum, because they are not publicly priced.

    We sell skills that Claude Code, Cursor, Codex CLI and Copilot load through one connection to Agensi (see the MCP setup guide), and we name a few of ours in the sections where they cut cost. No vendor paid to be listed.

    What do you pay for: a flat plan or tokens?

    Claude is sold two ways. A subscription charges a flat monthly fee for an allowance of usage. The API charges per million tokens (MTok), with one rate for what you send and a higher one for what the model writes back. Claude Code can be billed either way: it is included in the paid plans, and you can also point it at an API key. Our rule of thumb is simple. A person working by hand belongs on a plan. Anything that runs without a person belongs on the API. We have not measured where the two cross for your usage, so we do not claim a break-even point.

    What do the Claude plans cost?

    Read on the Claude pricing page on 09-11-2026, vendor-stated, excluding tax. The page says prices may change.

    PlanPriceClaude CodeCowork
    Free$0NoNot listed
    Pro$17 a month on annual billing ($200 up front), $20 monthlyYesNot stated
    Max$100 a month, 5x Pro usageYesNot stated
    Ultra$200 a month, 10x Pro weekly usageYesNot stated
    Team Standard seat$20 a month annual, $25 monthlyYesYes
    Team Premium seat$100 a month annual, $125 monthlyYesYes
    Enterprise$20 per seat a month billed annually, 20 seat minimum, plus usage at API ratesYesInherits Team features per the page

    When a plan hits its limit you can switch on usage credits. The support article on usage credits says they bill at standard API rates, that you set a monthly spend cap and that Claude Code draws from the same credits. Purchases are non-refundable, and credits bought on or after 07-12-2026 expire after 12 months. The monthly API credits that come with Max and Ultra ($100 and $200) are separate from usage credits. Our verdict: set the monthly cap on day one, because a cap you chose beats a bill you did not expect. Whether Cowork runs on Pro and Max is the one open question here, and our Claude Cowork pricing guide is where we track it.

    What does a task cost on the API?

    The rates behind every cost below come from the Anthropic API pricing page, read on 09-11-2026, per MTok.

    ModelInputOutputCache read5 minute cache write
    Claude Opus 5.5$4$20$0.20$5
    Claude Sonnet 5.5$2$10$0.10$2.50
    Claude Haiku 5.5, prompts up to 100K tokens$0.10$0.50$0.01$0.125

    The Batch API takes 50% off input and output, so Sonnet 5.5 becomes $1 and $5. Web search costs $10 per 1,000 searches plus the tokens of the results. Claude Fable 5.1 sits above these at $10 input and $50 output, and we leave it out of the scenarios. Claude 4.7 and later use a tokenizer that yields about 30% more tokens for the same text (vendor-stated), so a cost you calculated on an older model will run higher on a newer one.

    What does a solo developer pay?

    Task one, fixing one bug in a mid-size repo in Claude Code. Assumption, not measured by us: about 25 requests with 15,000 fresh input tokens, 60,000 cache write tokens, 1.2 million cache read tokens and 20,000 output tokens. Calculated by us: $0.50 on Sonnet 5.5 and $1.00 on Opus 5.5.

    Task two, adding a feature with tests in a two hour session. Assumption: about 150 requests with 100,000 fresh input, 400,000 cache write, 13.5 million cache read and 120,000 output tokens. Calculated by us: $3.75 on Sonnet 5.5 and $7.50 on Opus 5.5.

    For a full working day, we lean on the vendor rather than on our own guess. The Claude Code cost page says the average is about $13 per developer per active day and $150 to $250 per developer per month across enterprise deployments, with 90% of users below $30 per active day. Our judgment: for a solo developer, Pro is the cheapest sensible start, and the day you hit the session limit twice in one week is the day Max at $100 pays for itself in time saved. We have not tested how many sessions fit in a Pro allowance, so that threshold is a judgment, not a measurement. Developers who want the numbers for their own sessions can use Claude Code Token Meter, which by its listing audits exact token usage and installs a live meter, or TokenOps, which by its listing reads your local logs and reports tokens, estimated cost and limit reset times. We have not run either one.

    What does an API builder pay?

    A pull request review bot, 40 pull requests a week, 50,000 input and 3,000 output tokens each, no caching (assumption, not measured by us). Calculated by us: $0.13 per pull request on Sonnet 5.5, so $5.20 a week, $2.60 with the Batch API and $10.40 on Opus 5.5. Teams building this kind of bot should start in the code review skills.

    A support ticket classifier, 10,000 tickets a month at 600 input and 100 output tokens each (assumption). Calculated by us: $1.10 on Haiku 5.5, $0.55 with Batch, $22 on Sonnet 5.5 and $44 on Opus 5.5. That is a 40 times spread for a job that rarely needs Opus, and it is the reason we tell API builders to pick the smallest model that passes their test set. Our API and backend skills cover the plumbing around that choice, and our Claude API pricing guide goes deeper on billing and spend limits.

    A knowledge base of 50,000 tokens reused across 100 calls inside the cache lifetime, on Sonnet 5.5. Calculated by us: $10.00 without caching and $0.62 with it, one cache write plus 99 cache reads. The default cache lifetime is 5 minutes and each use refreshes it, so a break longer than that means a new write. Prompts need a minimum length to be cached: 512 tokens on Sonnet 5.5, Opus 5.5 and Haiku 5.5, per the prompt caching docs. Our prompt caching guide shows how to structure the prefix.

    What does a marketer pay?

    A marketer who runs a weekly content brief and report routine is priced by plan, not by token: Pro is $17 a month on annual billing or $20 monthly. We did not run such a routine, so we cannot tell you how many runs fit before the session limit message appears. Not tested by us. Run your own routine on a Pro account for a week, count the runs and the day the limit appears, and decide on that number. Marketing teams can browse the marketing and SEO skills that package such routines.

    What does a five-person dev team pay?

    Three routes, calculated by us from vendor prices. Team Standard seats: 5 x $25 = $125 a month, or $100 billed annually. Team Premium seats: 5 x $125 = $625 a month, or $500 annually. Pay per token at the vendor average of $150 to $250 per developer per month: $750 to $1,250 a month (our multiplication of the vendor range). Enterprise starts at 20 seats x $20 = $400 a month before usage, so it only applies once the team is larger. We would send a team of five to Team Standard first, and Premium only after someone proves the Standard allowance runs out mid week. What would change our mind: the vendor average describes API billed enterprise usage, and a seat has limits we have not measured, so a heavy team can still end up on credits. Whoever signs off the budget will find the finance and accounting skills useful for tracking the spend.

    What makes the bill grow?

    Four things, in the order we would check them. Long sessions, because every request re-reads the context, and a cache that expired after a break turns that read back into a full price write. Agent teams, which the cost page says use about 7 times the tokens of a standard session when teammates run in plan mode: applied to our bug fix that is about $3.50 on Sonnet 5.5, our calculation from an approximate vendor figure. Noisy tool output: a 30,000 token test log that stays in context for 40 more requests costs about $0.12 at the Sonnet 5.5 cache read price, against $0.002 if a filter reduces it to 500 tokens (calculated by us; the vendor lists hooks that filter output as a tip). And web search: 200 searches cost $2.00 before the tokens of the results. Researchers who run many searches should look at the research and analysis skills. We did not verify how extended thinking is billed on the current models, so read the extended thinking docs before you count on a number.

    How do you pay less?

    The cost page lists the levers, and we rank them by how often they apply. Match the model to the job, so a classifier does not run on Opus. Clear the context between unrelated tasks. Move long instructions out of CLAUDE.md into skills that load on demand. Disable MCP servers you do not use. Keep agent teams small and use Sonnet for teammates. Our fuller walkthrough sits in our guide to skills that reduce Claude Code costs.

    Skills that address this, each described as its listing describes it, none run by us. Context Rent finds what eats your context window, such as whole file reads, noisy commands, cache expiry and a large CLAUDE.md. TokenSaver compacts verbose tool output by about 90% according to its listing, a maintainer claim we have not verified. slim-context claims to cut context burn by 30 to 50% by flattening repository noise, again a claim we have not checked. Token Budget Architect claims to cut monthly API bills by 35 to 55% through prompt cache prefix stability, which is the same lever as our caching example above, and we have not measured its result. Teams that run this kind of tooling in production will want the LLM ops skills.

    Which route we send each profile to

    Solo developer: Pro, then Max once the session limit costs you more than $80 a month in lost time. Small dev team: Team Standard, annual billing if you are sure of the headcount. Marketer: Pro, with the usage credit cap set on day one. API builder: the smallest model that passes your test set, Batch for anything nobody waits on, caching for any fixed prefix above 512 tokens. Budget owner: ask for a month of logged usage from /usage or the Console before approving seats beyond Standard.

    Prices and limits change often, so recheck each figure against the vendor pages linked above before you put it in a budget. This is cost information, not financial advice, and you should compare the numbers with your own invoice. Our Claude Code usage limits guide is the next read if limits, not price, are what slow you down.

    Written by Samuel Rose, Founder of Agensi. Next scheduled review: 09-12-2026.