Claude Code Pricing (October 2026): Every Claude Plan, Seat and API Rate

TL;DR

As of October 2026, Claude costs $0 on Free, $20 a month on Pro ($17 a month billed annually), $100 on Max 5x and $200 on Max 20x, and $25 or $125 a seat on Team monthly billing ($20 or $100 billed annually). Enterprise is $20 a seat a month billed annually, with a 20 seat minimum, plus usage at API rates (claude.com/pricing). Claude Code is included in every paid plan and draws from the same usage pool as chat. On the API, Claude Opus 5.5, released on 22 September 2026, costs $4 per million input tokens and $20 per million output tokens; the previous Claude Opus 5 costs $5 and $25. Claude Sonnet 5.5, released on 28 September 2026, costs $2 and $10, the same as Sonnet 5, Claude Haiku 5.5, released on 7 October 2026, costs $0.10 and $0.50 for prompts up to 100,000 tokens ($0.50 and $2.50 above that) against $1 and $5 on Claude Haiku 4.5, and Claude Fable 5.1 costs $10 and $50. Subscriptions and the API meter the same thing underneath, which is tokens, so the line a team controls on its Claude Code bill is how many tokens each request carries. Sentra, the managed organization memory for teams and agents, comes into the cost picture only at that point, after the plan arithmetic below.

How much does Claude cost per month? Every plan compared

Claude costs between $0 and $200 a month for one person and between $20 and $125 a seat for a team. Anthropic sells Claude two ways: a subscription that bundles usage into a flat fee, and the API, which bills per token. Claude Code is included in every paid subscription tier. As of October 2026, source: claude.com/pricing. Every Claude plan, monthly and annual.

Two things in that table trip people up. Max is a personal plan and is billed monthly only, so there is no annual discount to negotiate (support.claude.com). And the Max multipliers apply per five-hour session rather than per month, with a weekly limit on top, so they raise the ceiling on a single long agentic run rather than granting a monthly pool you can spend however you like. On every paid plan, usage credits let you keep working past a limit at standard API rates (support.claude.com).

How much does Claude Enterprise cost?

Claude Enterprise costs US$20 per seat per month, billed annually, with a 20 seat minimum, plus usage billed at API rates, so the total scales with the models and tasks your team runs (claude.com/pricing). It is the only Claude plan where seats and usage are billed separately: the seat fee buys access and controls, and every token is metered at the same per-million rates as the API. Enterprise can be bought self-serve or through sales, and only the sales-assisted route adds invoicing on net terms, AWS Marketplace availability and tiered incentives on committed spend.

• Enterprise adds role-based access with fine-grained permissions and SCIM provisioning, neither of which Team includes.
• Enterprise adds audit logs and the Compliance API for observability and monitoring.
• Enterprise adds custom data retention controls, network-level access control and IP allowlisting.
• Enterprise offers a HIPAA-ready configuration and lets admins set user and organization spend limits on metered usage.

For comparison, OpenAI lists ChatGPT Business at $25 per user per month, or $20 billed annually (help.openai.com), and does not publish a ChatGPT Enterprise seat price; Enterprise is quoted by its sales team (chatgpt.com/pricing). Claude Enterprise is unusual in publishing its seat price. The decision rule: buy Enterprise when you need SCIM, audit logs, retention controls or HIPAA readiness, and budget its usage line from your expected tokens rather than from the seat price.

What does the Claude API cost per million tokens?

The Claude API costs $0.10 to $10 per million input tokens and $0.50 to $50 per million output tokens on the current models, with output always priced at five times input. As of October 2026, source: claude.com/pricing and Anthropic's pricing documentation. All prices are US dollars per million tokens.

Four modifiers change those rates. The Batch API takes 50% off input and output for asynchronous work. A one-hour cache write costs 2x the base input rate, against 1.25x for five minutes. US-only inference costs 1.1x on input and output. Fast mode for Opus 5.5 runs up to 2.5x faster at 2x standard pricing, which is $8 input and $40 output. Claude 4.6 and later models include the full 1M token context window at standard rates, except Haiku 5.5, which charges $0.50 input and $2.50 output once a prompt passes 100,000 tokens. The full model-by-model breakdown, including older models, is in our Claude API pricing guide.

Is Claude Team worth it compared to buying Pro seats individually?

On price alone, no. Team Standard costs $25 per seat monthly against $20 for Pro, and $20 per seat annually against $17. That is 25 percent more per person on monthly billing and 18 percent more annually, and the reason to pay it is administrative rather than economic: one invoice, centrally managed seats, shared projects, single sign-on, enterprise search, and the admin controls that make offboarding a click instead of a chase. Team Standard seats also carry more usage than Pro. Team has a two seat minimum and a 150 seat maximum. If those things are worth $3 to $5 per person per month to you, buy Team; if you are three developers who already expense your own subscriptions, Pro seats are cheaper and the difference is real.

Claude Code subscription vs API rates: which is cheaper?

This has a break-even you can compute rather than argue about. Take a substantial agentic request at roughly 20,000 input tokens and 1,500 output tokens. On Claude Opus 5.5, at $4 per million input and $20 per million output, that request costs $0.08 plus $0.03, about $0.11 on the API. On Claude Sonnet 5.5, at $2 and $10, it costs $0.04 plus $0.015, about $0.055. The table divides each monthly price by those request costs and spreads the result over 21 working days.

Read it as a threshold, not as a capacity promise, because every plan still has five-hour and weekly limits. A developer who makes more than about nine serious Opus 5.5 agent calls in a working day is already better off on a $20 subscription than on API billing, provided the work fits inside Pro's limits, which is why almost every full-time user of a coding agent should be on a plan rather than a key. The API wins in the opposite cases: bursty or automated workloads, anything you want to run in batch at half price, and any product you are building on top of Claude, where per-seat licensing does not apply. Both numbers assume the request shape above, so recompute with your own average if your context payloads are larger, and note that larger payloads move the break-even in favour of the subscription.

What is the cheapest way to reduce Claude token costs for a team?

Four levers exist, and they are not equally powerful. Ranked by what they actually do: route easy work to a smaller model, which is usually the largest single reduction available and the one teams resist longest, since Haiku 5.5 costs 97.5% less than Opus 5.5 on both input and output for prompts up to 100,000 tokens ($0.10 against $4 in, $0.50 against $20 out) and 87.5% less above that; turn on prompt caching, where a cache read costs 0.1x the base input rate on most models (0.05x on Opus 5.5 and Sonnet 5.5, 0.025x on Fable 5.1) against a write at 1.25x for five minutes or 2x for an hour, so a stable prefix gets about 90 percent cheaper to resend (Anthropic docs); use the Batch API for anything asynchronous, which is a flat 50 percent off both input and output; and send less context in the first place. The first three lower the price of the tokens you send. Only the fourth lowers how many you need to send, which is why it is the only one that keeps working as your corpus grows, and it is the lever the rest of this page is about.

How does Claude Code pricing scale for a growing team?

Claude Code pricing scales linearly with seats on Team, and with seats plus metered tokens on Enterprise. Sentra publishes this breakdown because the budgeting question arrives before the optimization question. At Anthropic's current seat prices (as of October 2026, source: claude.com/pricing: Team Standard $25 per seat monthly or $20 billed annually, Team Premium $125 monthly or $100 annually, both including Claude Code), a team's subscription floor scales like this.

Two honest caveats on that table. First, seats are the floor, not the bill: heavy agentic use runs into plan limits, and teams that overflow onto usage credits or API rates pay by the token, which is where context size dominates. Second, the per-seat price is fixed by Anthropic, so the only line a team controls is tokens per request. You can model your own fleet, including the effect of prompt caching and shared memory, with our agent token cost calculator at /tools/agent-token-cost-calculator.

What does Claude cost for one developer, a 10-person team and a 50-person team?

A single developer pays $17 to $200 a month, a 10-person team pays $200 to $1,250, and a 50-person team pays $1,000 to $6,250 in seats, before any metered usage. The worked examples below use the October 2026 prices above and one assumption about usage, which is 25 agentic requests per developer per working day at the 20,000 input and 1,500 output token shape, over 21 working days. Replace that assumption with your own logs.

• One developer on the API pays 25 × $0.11 × 21 = $57.75 a month on Opus 5.5, or 25 × $0.055 × 21 = $28.88 on Sonnet 5.5, so Pro at $20 is cheaper whenever that workload fits Pro's limits, and Max 5x at $100 is the step up when it does not.
• The same developer with prompt caching, where 15,000 of the 20,000 input tokens are Opus 5.5 cache reads at $0.20 per million, pays $0.02 + $0.003 + $0.03 = about $0.053 a request, or about $27.83 a month, roughly half the uncached API cost before cache writes.
• A 10-person team on Team Standard pays 10 × $20 = $200 a month billed annually, or 10 × $25 = $250 monthly, and a mix of eight Standard and two Premium seats pays (8 × $20) + (2 × $100) = $360 annually or $450 monthly.
• A 10-person team cannot buy Enterprise alone, because Enterprise has a 20 seat minimum. At the minimum it pays 20 × $20 = $400 a month in seats plus metered usage for the 10 people using it, and at the Sonnet 5.5 assumption that usage is 10 × 25 × $0.055 × 21 = $288.75, for a total of about $688.75 a month.
• A 50-person team on Team Standard pays 50 × $20 = $1,000 a month billed annually, and a mix of 40 Standard and 10 Premium seats pays (40 × $20) + (10 × $100) = $1,800.
• A 50-person team on Enterprise pays 50 × $20 = $1,000 a month in seats plus 50 × 25 × $0.055 × 21 = $1,443.75 of Sonnet 5.5 usage, about $2,443.75 in total, or about $3,887.50 if the same requests run on Opus 5.5 at $0.11 each.

The decision rule: Team is cheaper for most engineering teams because usage up to the seat limits is included, and Enterprise becomes worth its metered usage line when you need its controls, or when a small number of very heavy users would otherwise force the whole team onto Premium seats.

What Claude Code Pricing Actually Charges You For

Every request to Claude Code has two cost drivers: the tokens you send in and the tokens you get back. Most teams focus on the second one, trying to get shorter, cleaner outputs. But the input side is usually where the real spend hides, because input volume dwarfs output volume in agentic work.

That input is not just your question. It is the system prompt, the retrieved documents, the conversation history, the code snippets, the tool outputs, and whatever else got stuffed in to give the model a fighting chance at answering correctly. In agentic workflows especially, that context payload can outweigh the actual question by orders of magnitude. Enterprise AI agents that answer from company documents carry the same overhead on every call.

Model upgrades can also raise the input count without anyone sending more material. Anthropic's pricing documentation notes that Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text, so a team moving from Opus 4.6 at $5 per million input tokens to Opus 5.5 at $4 can pay 1.3 × $4 = $5.20 for text that used to cost $5. That is the usual answer when a bill goes up after a move to a model with a lower headline rate.

So when people ask how to reduce Claude Code costs, the honest answer is rarely "use a cheaper model." It is "stop sending the model so much raw material to sort through every single time."

Why Token Costs Spiral: The Re-Reading Problem

Most systems handle context with retrieval augmented generation. A query comes in, the system searches a vector store, pulls back a pile of chunks that seem relevant, and stuffs them into the prompt. The model then has to read all of it, figure out what actually matters, resolve any contradictions, and guess at relationships that were never made explicit.

This happens on every single query. The same documents get re-read, the same ambiguities get re-resolved, sometimes differently each time. Every chunk you retrieve is a chunk you pay to process, every time, forever. Scale that across thousands of queries a day and the token bill grows in direct proportion to how sloppy your context assembly is.

The Fix: Resolve Meaning Once, Not Every Query

The alternative is to stop treating context as something you rebuild at query time and start treating it as something you resolve once, upfront, and keep current.

This is the core idea behind Sentra. Instead of retrieving raw chunks and asking the model to make sense of them on the fly, Sentra resolves meaning at write time. Facts, relationships, and changes get processed into a bi-temporal context graph the moment they happen, not the moment someone asks a question about them. Bi-temporal means Sentra tracks both when something became true and when the system learned about it, so the graph knows the difference between what changed last week and what you just found out.

By the time a query arrives, the hard work is already done. The model gets a clean, resolved answer instead of raw material to sift through. Fewer tokens go in, because there is less noise to send. Fewer tokens get wasted on the model reasoning its way through contradictions, because the contradictions were already resolved before the question was ever asked.

What This Looks Like in Practice

On Terminal-Bench 2.1, Sentra's own internal evaluation (pending official verification), Codex CLI with GPT-5.5 xhigh and Sentra Code Memory scored 88.31% mean reward across 445 trials against the published 83.37% baseline, a gain of 4.94 points, with 72.6% lower model cost ($510.30 against $1,862.98) and 41.2% fewer tokens (results). The run used an OpenAI model, so it shows the mechanism rather than a Claude-specific saving: the agent spent fewer tokens rebuilding context it could read from memory.

That is the real lever for Claude Code cost optimization once model routing, caching and batching are in place. Prompt engineering tricks trim a request at the margin, while what you feed the model in the first place sets the size of every request.

FAQ

The Bottom Line

Claude pricing as of October 2026 is a flat fee for people and a meter for tokens: $20 for Pro, $100 or $200 for Max, $20 to $125 a seat for Team, and $20 a seat plus API-rate usage for Enterprise. Whichever plan you buy, the bill underneath is set by how many tokens each request carries, and retrieval-style context assembly makes you pay for the same unresolved material over and over. Sentra, the managed organization memory for teams and agents, resolves what your organization knows once and keeps it current, so agents read a small, correct answer instead of re-reading raw sources. Pick the plan from the tables above, then cut tokens per request, because that is the one line on the bill Anthropic does not set.