Claude Pricing in 2026: Every Plan and Model Rate, Checked Against Anthropic

Claude pricing in October 2026 runs from $0 to $200 a month for one person, from $40 a month for a 2-seat team, and from $0.10 to $10 per million input tokens on the API. We checked every line against Anthropic's own pages on 8 October. Two widely reported changes are real, neither of them happened in September, and the change that matters most for your bill shipped on 7 October.

Claude pricing in one table: every plan, checked

Claude costs $0 on Free, $20 a month on Pro ($17 billed annually), and $100 or $200 a month on Max. Team costs $20 to $25 per standard seat and $100 to $125 per premium seat, and Enterprise costs $20 per seat plus usage. API rates run from $0.10 to $10 per million input tokens, depending on model.

Those numbers come from three Anthropic surfaces that do not always agree on detail. The public Claude pricing page shows the plan cards. The Help Center articles carry the seat rules and the Max tier prices. The developer documentation carries every per-token rate, with footnotes that matter more than the table.

Prices are Anthropic's US web prices excluding tax. Mobile app pricing may differ by platform, per Anthropic's Max plan article. Usage multiples are Anthropic's own descriptions, not token counts.

The individual ladder is simple on price and vague on quantity. Pro is $20 a month, or $200 paid up front for a year. Max is $100 for 5 times Pro's usage and $200 for 20 times, both billed monthly only, according to Anthropic's Max plan article.

One small oddity is worth flagging for anyone quoting the pricing page. The Max card on claude.com shows only "From $100". The $200 figure for Max 20x lives in the Help Center, not on the card. If you are building an internal price list, cite the Help Center for the upper tier.

Usage is described in multiples, never in tokens. Limits reset on a rolling 5-hour session window, and paid plans add weekly limits on top. Anthropic also reserves the right to apply other caps. That is the single largest gap in the whole price list, and we come back to it below.

Team is now the plan for 2 to 150 people. You can mix standard and premium seats. A standard seat gets 1.25 times Pro's per-session allowance and a premium seat gets 6.25 times, which is the "5x more than standard" the pricing card describes.

Enterprise is a different animal. It is listed at $20 per seat plus all usage at API rates, billed annually. We covered that structure, and what it does to your exposure, in the analysis of Anthropic's move from bundled seats to metered usage. We will not repeat it here.

Claude API pricing in 2026: the per-token rates

The API is where nearly every price movement in the last four months has landed. The current family is Fable 5.1 at the top, then Opus 5.5, Sonnet 5.5 and Haiku 5.5. Older models stay available at their own rates, and some of them are now more expensive than their replacements.

Source: Anthropic pricing documentation and claude.com/pricing, both accessed 8 October 2026. Batch prices are half the standard rate, as the documentation states. Older Sonnet 4.x and Opus 4.x rates are omitted for space.

Caching, batch and the modifiers that move the real rate

The headline rate is rarely what you pay. Writing to the 5-minute prompt cache costs 1.25 times the base input price, and the 1-hour cache costs 2 times. Reading from cache normally costs 0.1 times input. On Opus 5.5 and Sonnet 5.5 it costs 0.05 times, and on Fable 5.1 it costs 0.025 times.

The Batch API takes 50% off input and output for asynchronous work. US-only inference adds a 1.1 times multiplier on Claude 4.6 and later models. Fast mode on Opus 5.5 costs $8 input and $40 output, double the standard rate. The full 1M-token context window bills at standard rates on every current model except Haiku 5.5, which charges more above 100K tokens.

Our view is that the cache-read cuts matter more than the headline cuts for anyone running agents. An agent rereads the same system prompt and tool definitions on every turn. On that workload the cache-read line, not the input line, is your bill.

The tokenizer catch inside the Sonnet price

Per-token prices hide a unit change. Anthropic's documentation says Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text. The Sonnet 5 launch post puts the range at roughly 1.0 to 1.35 times, depending on content.

So run the arithmetic yourself. Text that cost $1.00 on Sonnet 4.6 at $3/$15 costs about $0.67 on Sonnet 5 at the same token count. Stretch the token count by 1.35 and it costs about $0.90. That is our arithmetic, not Anthropic's, and it means the real saving on identical text sits somewhere between 10% and 33%, not a flat third. The broader pattern, where cheaper tokens still produce larger invoices, is the subject of the piece on why falling token prices raise AI bills.

What actually changed in Claude pricing since June

The brief for this post said two things changed in September: the Team minimum dropped and Sonnet's introductory rate became permanent. Both changes are real. Neither is dated to September on Anthropic's own record. Here is the sequence as the primary sources show it.

Sonnet kept $2/$10, and the increase is gone

Sonnet 5 launched on 30 June at $2 per million input tokens and $10 per million output tokens. That rate was labelled introductory through 31 August. On 10 August, Anthropic edited the Sonnet 5 announcement to say the rate was now permanent and the $3/$15 standard pricing would no longer apply.

The pricing documentation carries the same statement as a footnote. Its wording is that the scheduled increase will not occur. That is stronger than a postponement, and we read it as a commitment rather than a promotion that might return.

Claude Team plan pricing now starts at 2 seats

When Anthropic launched Team in May 2024, it cost $30 per user per month with a minimum of 5 seats. Today the Help Center states that Team plans require a minimum of two members, and the seat price is $25 monthly or $20 billed annually.

At today's prices, the old 5-seat floor would cost $100 a month on annual billing. The new 2-seat floor costs $40. The per-seat price did not move with the minimum. What moved is who can buy the plan at all, which is the 2 to 4 person company that was previously stacking individual Pro accounts.

The 5.5 family cut rates at the top and the bottom

Opus 5.5 arrived on 22 September at $4/$20, which Anthropic describes as 20% less than Opus 5 per token. Anthropic also claims it costs 40% less to run on typical workloads, because it uses fewer tokens per task. Haiku 5.5 arrived on 7 October at $0.10/$0.50 for prompts up to 100K tokens, 90% below Haiku 4.5 on that band.

Sonnet 5.5 is the exception that proves the pattern. Its list price did not move from Sonnet 5. Its cache reads were halved to $0.10 a week later, which on agent workloads is a price cut wearing a different label.

The Claude pricing change hiding in the release notes: API credits on Max and Team

The most consequential change of the quarter got one line in the Haiku 5.5 announcement. From 7 October, Max and Team subscribers receive a monthly credit for the Claude API, claimed by linking a Claude Console organisation to the subscription.

The amounts are not random. Per Anthropic's API credits article, Max 5x gets $100 a month and Max 20x gets $200. Team gets $20 per standard seat and $100 per premium seat, pooled across the organisation and capped at $500. Each credit matches the seat's annual-billing price exactly.

The rules that decide whether the credit is worth anything

The fine print is where this either pays or evaporates. Credits expire at the end of each billing cycle. They are spent before any credits you bought. You can claim only after 7 days on an eligible plan, and you can link only one Console organisation.

The exclusions matter too. The credits do not cover interactive Claude Code, extra usage on the chat plan, or Claude running through Bedrock, Vertex or Foundry. Free, Pro and Enterprise plans are not eligible at all. If your team runs its API work through a cloud marketplace, this credit does nothing for you.

Claude Max plan or API: where the comparison breaks

The question buyers ask most is whether Max is cheaper than paying the API directly. The honest answer is that nobody outside Anthropic can compute it, because Anthropic publishes Max limits as multiples of Pro, not as tokens. You cannot price a token allowance you cannot see.

That gap is not unique to Anthropic. It is the industry norm, and we have argued it should end in the case for publishing rate limits as numbers. A plan that quietly changes its multiples is also a plan that can shrink without changing its price, the pattern traced in the analysis of AI usage-limit shrinkflation.

The credits change the comparison anyway. If you already spend $100 a month on direct API calls, Max 5x now costs you nothing extra for the chat plan, because its $100 credit covers the API spend you were making. Our position is simple. For any developer already spending $100 to $200 a month on the API outside a cloud marketplace, a Max plan is now the cheaper way to buy those tokens.

The same logic decides the small-team question. Two Pro accounts on annual billing cost $34 a month. A 2-seat Team on annual billing costs $40, carries 25% more usage per seat, adds admin controls, and returns $40 a month of API credit. For any company paying for two or more people, we would no longer buy individual Pro accounts.

What the Claude pricing changes signal about seats versus usage

Read the quarter as one decision, not six announcements. Anthropic held its subscription price points and spent its pricing effort in two places: lower per-token rates on new models, and more value attached to each seat. The seat price is the anchor. Everything around it moved.

The credit is the clearest signal. A credit equal to the seat price, which expires monthly and requires linking a developer console, is not really a discount. It is a conversion device. It turns every Max subscriber and every small Team admin into an API account holder with a funded balance and a reason to build something before the month ends.

The $500 pooled cap tells you who the gift is for. Twenty-five standard seats, or five premium seats, exhaust it. A large company gets almost nothing per head. A 2 to 20 person company gets its subscription back. That is a deliberate tilt toward the smallest buyers, and it matches the cut to a 2-seat Team minimum.

We think this is the right strategy for Anthropic and a mixed one for you. Usage is where its revenue grows and where switching gets harder, because prompts, caches and evals get tuned to one model family. The credit makes that tuning free to start, which is exactly why it is worth watching. The lock-in mechanics are covered in the breakdown of what switching model providers really costs.

The per-token cuts point the same way. Opus 5.5, Sonnet 5.5 and Haiku 5.5 each come with a claim about lower cost per task: 40%, up to 30%, and around 75% respectively. All three are Anthropic's own testing. They push buyers to compare models on task cost rather than token price, which is the right frame, and one that only you can measure on your own workload. The case for routing work across tiers is set out in the analysis of model routing and its quality gap.

Where this reading of Claude pricing is weakest

Three parts of this post rest on thinner evidence than the price tables, and you should weigh them accordingly.

The Team date is not on Anthropic's record

We could verify the current 2-seat minimum, and the 2024 launch post verifies the old 5-seat floor. We could not find a dated Anthropic source for the change itself. The Help Center release notes, which go back to August 2025, carry no entry for it. Press coverage dates it to July 2026, and we could not open those articles during research. Treat the timing as reported, not confirmed.

A week-old credit is a promotion until it survives a renewal

The API credit launched on 7 October and was still rolling out when we checked. Anthropic's pricing page says plans and prices are subject to change at its discretion. A benefit this generous, matched exactly to the seat price, could be trimmed, capped harder or limited to certain models. Our Max-versus-API conclusion holds only while the credit holds.

Cost-per-task claims are vendor numbers

The 40%, 30% and 75% figures come from Anthropic's own tests, with no published task list or method on the announcement pages. Per-token prices are facts. Per-task savings are claims. Your own workload, measured before and after a switch, is the only number that should change a budget. A mixed-tier deployment is also harder to forecast, because the conversion between credits, tokens and dollars differs at every layer.

Frequently asked questions

Where to start this week

If your company pays for two or more individual Pro accounts, price a 2-seat Team on annual billing today. It costs $6 a month more than two annual Pro plans and returns $40 of API credit, which most teams will use within the first week.

If you already hold a Max or Team plan, link a Console organisation once your 7-day wait is over, then point one real workload at the credit before the cycle ends. Measure what that workload costs per task on Sonnet 5.5 and on Haiku 5.5. That one comparison will tell you more about your Claude bill than any price table, including this one.

Related reading