Claude Code API Pricing vs Plans: Anthropic, Sept 2026
Claude Pro, at $20 a month billed monthly, is the cheapest plan with Claude Code. At Anthropic's example session of 50,000 input and 15,000 output tokens, API list rates cost less than Pro below 40 Opus 5.5 sessions a month.
Claude Pro is the cheapest plan that includes Claude Code: $20 a month billed monthly, or $17 a month on a $200 annual payment.1 Anthropic lists both figures before tax, and we checked them on 28 September 2026.1 Through the API, Claude Code has no plan fee and bills per million tokens (MTok), from $1 input and $5 output on Haiku 4.5 up to $10 and $50 on Fable 5.1.1 At the token volume of Anthropic’s own one-hour coding-session example, Pro’s $20 equals the API cost of 40 sessions on Opus 5.5.1,2
TL;DR
- Four priced routes include Claude Code. Billed monthly and before tax, they are Pro at $20, a Team Standard seat at $25, Max from $100 and a Team Premium seat at $125.1 Annual billing cuts Pro to $17 and the Team seats to $20 and $100. Max bills monthly only.1
- On the API, Opus 5.5 costs $4 input and $20 output per MTok. Anthropic calls it the “Daily driver for agentic coding and enterprise work”.1
- A session of 50,000 input and 15,000 output tokens, the size of Anthropic’s worked example, costs $0.50 on Opus 5.5 and $0.25 on Sonnet 5 at list rates.2 Below 40 Opus 5.5 sessions a month, the API costs less than Pro.
- Above the break-even, a plan wins only if its usage limit covers the work. Anthropic states plan limits as multiples of other plans and gives no token count.1
- Enterprise charges $20 a seat a month plus usage at API rates. For the same tokens it costs $20 a seat a month more than the API alone, in exchange for administration controls.1
- Claude Code pricing sits inside our model infrastructure coverage, next to the wider Claude API pricing research.
How this ranking was built
Glamdring Research ranked every paid Claude plan or seat that includes Claude Code by one field: the listed price per user per month when billed monthly, in US dollars.1 The source is Anthropic’s plans and pricing page, captured on 28 September 2026. It states: “Prices shown don’t include applicable tax.”1
The page lists seven routes: Free, Pro, Max, the Team plan’s Standard and Premium seats, Enterprise and an Education plan.1 Free drops out because its comparison table marks Claude Code “No”.1 Education shows no price. Enterprise bills annually and adds usage at API rates, so it has no monthly-billed figure.1 Four routes remain. Max is ranked at its “From $100” entry price, because the page prints no separate prices for its 5x and 20x options.1
Monthly billing is the ranking field because Max has no annual option.1 API rates come from the same page and from Anthropic’s pricing documentation, which states “All prices are in USD.”2 Session costs and break-even counts are our arithmetic on those listed rates, with the inputs shown beside each figure. Our research methodology sets out how company-published prices are handled.
Which Claude plans include Claude Code, and what do they cost?
Claude Pro is the cheapest plan with Claude Code, at $20 per user per month billed monthly.1 A Team Standard seat costs $25, Max starts at $100 and a Team Premium seat costs $125, all billed monthly and before tax.1 Anthropic’s pricing FAQ states that Claude Code “is included in all paid plans”.1
| Rank | Plan or seat | Billed monthly (USD per user per month, before tax) | Billed annually (USD per user per month) | Usage as Anthropic states it |
|---|---|---|---|---|
| 1 | Pro | $20 | $17 ($200 billed up front) | At least 5x Free per 5-hour session |
| 2 | Team Standard seat | $25 | $20 | More than Pro |
| 3 | Max | From $100 | Not offered | 5x or 20x Pro per 5-hour session |
| 4 | Team Premium seat | $125 | $100 | 5x a Standard seat |
| Unranked | Enterprise | Not offered | $20 per seat, plus usage at API rates | Usage billed at API rates |
| Unranked | API through a Console account | No plan fee | No plan fee | Per-token rates |
| Excluded | Free | $0 | Not applicable | Claude Code not included |
Every price and usage description is Anthropic’s, as the page gives it.1
Every plan draws Claude Code from the same allowance as chat. Anthropic’s FAQ says Claude Code “shares the same usage limits as the rest of your plan, so your work in the terminal and your chats draw from one pool.”1 Those limits reset on a rolling five-hour window, and paid plans add weekly limits on top.1
Annual billing saves a fixed amount. Pro costs $240 a year billed monthly against $200 paid up front, a $40 saving. A Team Standard seat costs $300 a year monthly against $240 annually. A Team Premium seat costs $1,500 against $1,200.1 Max has no annual option, and the FAQ confirms both Max options are “billed monthly”.1
The Team plan is sold “For teams of 2 to 150”, so the smallest Team bill is two Standard seats at $50 a month billed monthly.1 Seat types can be mixed. The plan’s feature list includes “Claude Code and Claude Cowork”, and its comparison table marks Claude Code “Yes” for Team with no split by seat.1 A Premium seat billed monthly costs $25 more than Max’s entry price. It adds central billing, single sign-on and admin controls.1 The two capacities cannot be compared on the page, because a Premium seat is defined against a Standard seat and Max against Pro.
Model access differs by plan. Opus is available on Pro, Max and Team. Fable is reached through usage credits on Pro and capped at “50% of weekly limits” on Max.1
Enterprise is priced at “US$20/seat/month, billed annually”, or $240 a seat a year, with “Usage cost scales with model and task.”1 That structure makes Enterprise the API plus a seat fee. It buys controls the other plans lack, such as SCIM, audit logs and custom data retention.1
What are the Claude API per-token rates for Claude Code?
Claude Code on the API pays Anthropic’s standard per-token rates: $4 input and $20 output per MTok on Opus 5.5, $2 and $10 on Sonnet 5, $1 and $5 on Haiku 4.5, and $10 and $50 on Fable 5.1.1 On all four current models the output rate is five times the input rate. Anthropic’s documentation states every price in US dollars.2
| Model | Anthropic’s description | Input (USD/MTok) | Output (USD/MTok) | Cache write, 5-minute (USD/MTok) | Cache read (USD/MTok) | Batch input (USD/MTok) | Batch output (USD/MTok) |
|---|---|---|---|---|---|---|---|
| Fable 5.1 | “Next generation intelligence for long-running agents” | $10 | $50 | $12.50 | $0.25 | $5 | $25 |
| Opus 5.5 | “Daily driver for agentic coding and enterprise work” | $4 | $20 | $5 | $0.20 | $2 | $10 |
| Sonnet 5 | “High-performance model for coding and agents” | $2 | $10 | $2.50 | $0.20 | $1 | $5 |
| Haiku 4.5 | “Fastest, most cost-efficient model” | $1 | $5 | $1.25 | $0.10 | $0.50 | $2.50 |
List and cache rates are from Anthropic’s pricing page; batch rates are from its documentation.1,2
Caching changes the input side of the bill. A 5-minute cache write costs 1.25 times the input rate and a 1-hour write costs twice it.2 A cache read costs 10 per cent of input on most models, 5 per cent on Opus 5.5 and 2.5 per cent on Fable 5.1.2 Anthropic’s documentation says a 5-minute write pays for itself after one read, and a 1-hour write after two.2 Where a session resends the same files and instructions on each turn, the cached part is billed at the read rate.
The Batch API halves input and output prices.2 It processes requests asynchronously, so the discount fits queued jobs. A live, turn-by-turn terminal session is billed at the standard rates in the first four columns.
Three modifiers raise the rate. US-only inference costs 1.1 times standard on all token categories for Claude 4.6 and later models.2 Fast mode for Opus 5.5 costs $8 input and $40 output per MTok.2 Tools add input tokens: the tool-use system prompt adds 286 tokens on Opus 5.5, and web search costs $10 per 1,000 searches on top of tokens.2
Is Claude Code cheaper through the API or a plan?
Glamdring Research puts the break-even at 40 Opus 5.5 sessions a month against Pro’s $20, using the 50,000 input and 15,000 output tokens of Anthropic’s one-hour coding-session example.2 Below that count, the API at list rates costs less than Pro. Above it, Pro costs less only if its usage limit covers the extra sessions. Anthropic states that limit as a multiple of other plans and gives no token figure.1
The session cost comes first. Anthropic’s example uses 50,000 input tokens and 15,000 output tokens, with a cached variant in which 40,000 of the input tokens are cache reads.2 Priced at the current models’ list rates, the same session costs:
| Model | Session cost, no caching (USD) | Session cost, 40,000 cache reads (USD) |
|---|---|---|
| Fable 5.1 | $1.25 | $0.86 |
| Opus 5.5 | $0.50 | $0.348 |
| Sonnet 5 | $0.25 | $0.178 |
| Haiku 4.5 | $0.125 | $0.089 |
Our arithmetic: input tokens times the input rate plus 15,000 output tokens times the output rate. The cached column splits input into 10,000 uncached tokens and 40,000 cache reads.1,2 It leaves out the first cache write, as Anthropic’s example does. Writing those 40,000 tokens to the 5-minute cache on Opus 5.5 costs $0.20 once.1 Anthropic’s example also adds $0.08 of session runtime, a charge for its Managed Agents product that the figures above exclude.2
Dividing each plan price by a session cost gives the monthly session count at which the plan and the API cost the same:
| Route | Monthly price (USD) | Opus 5.5, no caching | Opus 5.5, cached | Sonnet 5, no caching | Sonnet 5, cached |
|---|---|---|---|---|---|
| Pro, billed annually | $17 | 34 | 48.9 | 68 | 95.5 |
| Pro, billed monthly | $20 | 40 | 57.5 | 80 | 112.4 |
| Team Standard seat, billed monthly | $25 | 50 | 71.8 | 100 | 140.4 |
| Max, entry price | $100 | 200 | 287.4 | 400 | 561.8 |
| Team Premium seat, billed monthly | $125 | 250 | 359.2 | 500 | 702.2 |
Prices from Anthropic’s pricing page; session costs from the table above.1
The decision turns on the model. Moving from Opus 5.5 to Sonnet 5 roughly doubles every break-even, because Sonnet 5’s input and output rates are half of Opus 5.5’s.1 At 20 working days a month, our assumption, Pro’s break-even on Opus 5.5 is two sessions a day. Max’s entry price covers ten.
Output tokens dominate the session bill. In an uncached Opus 5.5 session, output is $0.30 of the $0.50. With caching, output is $0.30 of $0.348, or 86 per cent.1 Caching cuts only the input side, so a workload heavy in generated code gains less from it than one that rereads a large codebase.
The constraint is the plan limit. Anthropic says “there’s no fixed message count” and that usage depends on conversation length, model and features.1 A plan caps spend at a known price. The API charges exactly for tokens used and has no plan cap. Run one real week of Claude Code on API credits, read the token count in the Claude Console and price it with the rates above.2 That week’s figure places the workload on one side of the break-even. Our explainer on what controls AI inference cost covers why token counts per task vary.
Why do other published Claude Code prices disagree?
Guides from Jet Admin and Superblocks disagree with Anthropic’s captured page on three points: the Max 20x price, the current Opus rate and Team seat access.3,4 Both guides list Max 20x at $200 a month.3,4 Anthropic’s page, captured on 28 September 2026, shows only “From $100” for Max.1
Jet Admin’s guide gives Opus 5 at $5 input and $25 output as a base rate.3 Anthropic now lists Opus 5 under Legacy models at that price and lists Opus 5.5 at $4 and $20.1 A guide that treats Opus 5 as the current rate prices Opus input and output 25 per cent above Opus 5.5.
Team access is the third point. Jet Admin describes Claude Code as sitting with Premium seats, and Superblocks prices Team at “$100-$125/seat/month”.3,4 Anthropic’s comparison table marks Claude Code “Yes” for the Team plan with no seat distinction.1 Confirm seat-level access in the current Team checkout before buying Standard seats for developers.
Third-party token allowances also circulate. Jet Admin states that Pro users get “approximately 44,000 tokens per 5-hour period”.3 Anthropic’s own FAQ gives no token figure and says usage depends on conversation length, model and features.1
Rate comparisons across model generations need one more adjustment. Claude 4.7 and later models use a tokenizer that “produces approximately 30% more tokens for the same text”, while Sonnet 4.6 and earlier use the previous one.2 At 30 per cent more tokens, Sonnet 5’s $2 input rate works out near $2.60 for text that Sonnet 4.6 bills at $3.1
What the ranking cannot tell you
The ranking prices access, and Anthropic does not publish the token capacity behind each plan. Pro, Max and Team usage is stated only as multiples: Pro at least 5x Free, Max 5x or 20x Pro, and a Premium seat 5x a Standard seat.1 The break-even counts show where the API stops being cheaper. They cannot show whether a plan’s limit reaches that far.
The prices are Anthropic’s own statements in US dollars before tax.1 Local-currency prices were not captured. The page also states that “Price and plans are subject to change at Anthropic’s discretion.”1 Volume discounts on the API are “negotiated on a case-by-case basis”.2
The session example is one size. Anthropic’s documentation puts one token at about 4 characters or 0.75 English words, and real sessions vary with repository size, model choice and tool use.2 Claude models on Amazon Bedrock and Google Cloud are priced by those providers and are outside this ranking.2
Frequently asked questions
Does Claude Code have a free API?
No free route includes Claude Code on a continuing basis. The Free plan’s comparison table marks Claude Code “No”.1 On the API, Anthropic’s documentation says “New users receive a small amount of free credits to test the API” and does not state the amount.2
Is Claude Code a subscription or an API product?
Claude Code is available both ways. Every paid plan includes it within the plan’s usage limits.1 Anthropic’s FAQ adds that “For heavy coding sessions, you can also switch to pay-as-you-go API credits through a Console account.”1 API billing is “based on actual monthly usage”.2
Why is the Claude API so expensive for coding sessions?
The API bills every token, and output costs five times input on all four current models.1 In a session of 50,000 input and 15,000 output tokens on Opus 5.5, output accounts for $0.30 of the $0.50 total.1,2 Caching and batch pricing reduce input and queued work. Neither reduces the output rate of a live session.
Can Claude Max be billed annually?
No. Anthropic’s comparison table lists Max 5x and Max 20x with a “Monthly” billing cycle, and its FAQ says both options are “billed monthly”.1 Pro and Team offer annual billing.1
Does US-only inference change what Claude Code costs on the API?
US-only inference multiplies every token price by 1.1 on Claude 4.6 and later models, including cache writes and reads.2 Opus 5.5 then costs $4.40 input and $22 output per MTok, and the 50,000-input, 15,000-output session rises from $0.50 to $0.55.1 Against Pro’s $20, the break-even falls from 40 sessions to 36.4.
Sources checked
Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.
