Claude Pricing 2026: Free, Pro, Max, Team and API Costs Explained
What Claude costs in 2026: Free, Pro at $20, Max at $100 or $200, Team and Enterprise, the five hour and weekly limits, Claude Code sharing, and API rates per million tokens. Checked 23 Sep 2026.
Claude has five plans. Free is $0. Pro is $20 a month, or $17 a month when you pay $200 for a year up front. Max starts at $100 a month for 5x the usage of Pro, and $200 a month for 20x. Team covers 2 to 150 people and Enterprise is priced on request. Every consumer plan uses a 200K context window and none of them includes API usage. The API is billed separately per million tokens: Sonnet 5 costs $2 in and $10 out, Opus 5 $5 and $25, Fable 5.1 $10 and $50, Haiku 4.5 $1 and $5. Usage limits on the subscriptions are relative, a five hour session cap plus a weekly cap, not a fixed number of messages.
Key Takeaways
- Pro is $20 a month or $200 a year ($17 a month). The annual plan is one payment, not twelve.
- Max is $100 a month for 5x Pro usage or $200 a month for 20x. Max also gets higher output limits, early access and priority at busy times.
- Limits are relative. Anthropic publishes a five hour session limit and a weekly limit, and Claude Code draws from the same pool as the chat app. There is no guaranteed message count on any plan.
- A subscription never includes the API. If you want to call Claude from code you pay per token on a separate Console account, from $1 per million input tokens (Haiku 4.5) to $10 (Fable 5.1).
- Sonnet 5 stays at $2 in and $10 out. Anthropic has said the planned rise to $3 and $15 on 1 September 2026 will not happen.
- Krater runs the same Claude models on a credit meter. In our three task run, a Sonnet 5 answer cost 0.6 to 3 credits, an Opus 5 answer 1.8 to 15 credits and a Fable 5.1 answer 3.3 to 28.8 credits, out of 1,500 credits on the $20 Pro plan.
Claude plans at a glance
| Plan | Price | Usage | What is added |
|---|---|---|---|
| Free | $0 | Base limit, resets every five hours | Web, iOS, Android and desktop apps, writing and editing, web search, memory, file creation and code execution, desktop extensions, Slack and Google Workspace connectors, remote MCP connectors, extended thinking, voice mode |
| Pro | $20 a month, or $200 a year ($17 a month) | More than Free, five hour sessions plus a weekly cap | Claude Code, Claude Cowork, Claude Design, Claude Science, unlimited projects, more Claude models, Claude for Microsoft 365, Claude in Chrome |
| Max 5x | $100 a month | 5x Pro | Higher output limits, early access to new features, priority at high traffic |
| Max 20x | $200 a month | 20x Pro | Same as Max 5x with four times the headroom |
| Team | Per seat, 2 to 150 users | Per member | Central billing and admin. Standard and Premium seats, see below |
| Enterprise | Custom | Custom | Spend limits, role based access, SCIM, audit logs, Compliance API, custom retention, network access control, IP allowlisting, HIPAA ready configuration |
Source: claude.com/pricing, read 23 September 2026. The pricing table there lists a 200K context window on every plan and shows model training as opt out on every plan.
What the $20 Pro plan actually buys
Pro is the plan most people mean when they search for Claude pricing. Anthropic's own description is $20 per month in the US, with the $17 figure only available if you pay $200 for the full year in one go. Three things about the plan matter more than the price.
The limit is a five hour session plus a weekly cap
Anthropic writes that the Pro session based usage limit will reset every five hours, and that Pro plans also have a weekly usage limit that applies across all models. The weekly limit resets at a fixed time each week and is tracked separately for Opus and for all other models. Long conversations, large file uploads and extended thinking all burn the session faster. There is no published message count, so any page that promises you an exact number of messages on Pro is guessing. We have a longer explainer at Claude usage limits explained.
Claude Code shares the same pool
Claude Code is included in Pro, which is a good deal if you code. The catch is that Anthropic states the limits are shared across Claude and Claude Code. A heavy coding session in the terminal uses up the same five hour window as your chats. When you run out, Anthropic's listed options are to upgrade to Max, turn on usage credits, switch to pay as you go API billing through the Console, or wait for the reset.
The subscription does not include the API
The Pro article says it plainly: The Pro plan does not include API usage. A Pro subscription and an API key are two separate accounts with two separate bills. If you want Claude inside your own app, script or automation, the next section is the one that applies.
Max: 5x or 20x for $100 or $200
Max is the plan for people who hit the Pro wall every day. On the pricing page it is listed as from $100 per month, and you choose between 5x and 20x the usage of Pro. Both Max tiers add higher output limits per response, early access to new features and priority access when Anthropic's servers are busy. The comparison table on the page also notes that the newest Fable models are capped at 50 percent of weekly limits on Max, while Pro reaches them through usage credits. Anthropic's dedicated Max help article returned an error when we checked it, so the figures above come from the pricing page itself.
Team and Enterprise
Team is sold per seat for groups of 2 to 150 and comes in Standard and Premium seats, with Premium including Claude Code. Anthropic shows the seat prices on the pricing page after you pick the Team tab, and we did not capture that table in this check. Third party pages we read on 23 September quote Standard at $25 a seat monthly or $20 on annual billing, and Premium anywhere between $100 and $150 a seat depending on the page and its date, so treat the Premium figure as unconfirmed and check the live page before you budget. Enterprise is custom priced and adds the governance features in the table above. Both business tiers have a minimum seat count, which several third party pages put at five for Team.
Claude API pricing per million tokens
| Model | Input per 1M tokens | Output per 1M tokens | Note |
|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | Current flagship. Cache hits $0.25 per 1M |
| Claude Fable 5 | $10 | $50 | Cache hits $1 per 1M |
| Claude Opus 5 (and Opus 4.5 to 4.8) | $5 | $25 | |
| Claude Sonnet 5 | $2 | $10 | Introductory price made permanent, the rise to $3 and $15 on 1 September 2026 was cancelled |
| Claude Sonnet 4.5 and 4.6 | $3 | $15 | |
| Claude Haiku 4.5 | $1 | $5 | Cheapest current model |
| Claude Opus 4 and 4.1 (retired) | $15 | $75 | Legacy pricing |
Source: docs.anthropic.com pricing, read 23 September 2026. Three modifiers on that page change the bill: prompt caching lowers the input price for repeated context, the Batch API has its own discounted rate for jobs that can wait, and regional or multi region endpoints on Bedrock and Google Cloud carry a 10 percent premium over the global endpoint. The same page warns that the tokenizer used from Claude 4.7 onward produces roughly 30 percent more tokens for the same text, so a prompt that cost X on an older model costs more than X on a newer one even at the same list price.
Subscription or API: which costs less for you
There is no single crossover point because the subscriptions have no published message count, but the shape is clear.
- If you chat with Claude yourself, in the app, most days, Pro at $20 is almost always cheaper than paying per token for the same conversations, and Max exists for when Pro's window is too small.
- If you want Claude inside a product, a workflow or a script, the API is the only route. Budget with the table above and remember output tokens are five times the input price on every model.
- If you mostly use Claude Code, Pro or Max is usually cheaper than API billing until the day you exhaust the shared limit, which is exactly what Anthropic's own support article describes.
- If you use Claude for some tasks but also want GPT, Gemini, image or video models, you are looking at two or three subscriptions, which is where a multi model workspace starts to make sense.
The same Claude models on Krater, and what they cost in credits
Krater gives you Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5 and the older Claude 4.x models inside one workspace next to 400+ other models, and bills every answer in credits based on what the model actually used. To show what that looks like in practice, we ran the three current Claude models side by side in Krater's Model Arena on 23 September 2026 on three ordinary business tasks: a 150 word client email, a Google Sheets SUMIFS formula with an explanation, and a 600 word blog outline plus a 250 word introduction. These are the credits Krater billed for each answer.
| Task | Claude Sonnet 5 | Claude Opus 5 | Claude Fable 5.1 | Whole three model round |
|---|---|---|---|---|
| 150 word client email | 0.6 credits | 1.8 credits | 3.3 credits | 6 credits |
| SUMIFS formula plus explanation | 0.9 credits | 4.7 credits | 5.6 credits | 12 credits |
| 600 word outline plus 250 word intro | 3.0 credits | 15.0 credits | 28.8 credits | 47 credits |
Three tasks is a cost sample, not a quality benchmark, and we are not ranking the models here. What it shows is the scale: on the $20 Krater Pro plan with 1,500 credits, everyday Sonnet 5 work costs well under one credit an answer, and even a long Fable 5.1 draft is about 2 percent of the month's credits. The Model Arena round itself, three flagship answers to the same prompt, cost 6 to 47 credits.
| Krater plan | Price | Credits a month | Context window | API access |
|---|---|---|---|---|
| Pro | $20 a month or $200 a year | 1,500 | 128K | No |
| Ultra | $49 a month or $490 a year | 4,000 | 200K | Yes, OpenAI compatible endpoints |
| Max | $119 a month or $1,190 a year | 10,000 | Up to 1M on supported models | Yes |
| Enterprise | Custom | Custom | Custom | Yes, plus SSO, SCIM, EU hosting, DPA |
Source: krater.ai/pricing. Two honest caveats. Krater is not a cheaper Claude: the credit meter tracks real model usage, so a heavy Fable 5.1 user could spend a Pro month's credits faster than they would hit Claude Pro's weekly cap, and the Krater Pro context window is 128K where Claude gives 200K on every plan. And Krater does not include Claude Code, Cowork, Design or the Chrome and Microsoft 365 integrations, which are Anthropic products. What you get instead is the ability to run Claude next to GPT, Gemini, Grok and DeepSeek in the same chat, compare them blind in Model Arena, and use image, video, voice, slides, research and document tools on the same credit balance, plus API access from Ultra at $49.
Which one to pick
- Choose Claude Free if you want to try Claude's models with web search, memory and file creation and can live with the base limit.
- Choose Claude Pro at $20 if Claude is your main assistant and you want Claude Code, projects and the desktop and browser integrations. Pay the $200 annual if you are sure you will use it for a year.
- Choose Claude Max if you hit Pro's five hour window most days, especially through Claude Code. Start at 5x and only move to 20x if you exhaust it.
- Choose the Claude API if you are building something. Prototype on Sonnet 5 or Haiku 4.5, measure your tokens, then decide whether Opus 5 or Fable 5.1 earns its price on your task.
- Choose Krater Pro at $20 if you want the same Claude models but also GPT, Gemini and 400+ others, plus image, video and voice tools, in one subscription. Choose Ultra at $49 if you also need an API key.
Frequently Asked Questions
How much does Claude cost per month?
Claude is free at the base tier. Pro costs $20 a month, or $17 a month when you pay $200 for a year in advance. Max costs $100 a month for 5x Pro usage or $200 a month for 20x. Team and Enterprise are priced per seat, with Team covering 2 to 150 users.
Is Claude Pro worth $20?
If you use Claude most days it usually is: you get more usage than Free, the newest models, Claude Code, unlimited projects and the Chrome and Microsoft 365 integrations. If you only chat a few times a week, Free covers most of what Pro does apart from the higher limit and Claude Code.
Does Claude Pro include API access?
No. Anthropic states that the Pro plan does not include API usage. API calls are billed per million tokens on a separate Console account, from $1 input and $5 output for Haiku 4.5 up to $10 input and $50 output for Fable 5.1.
How many messages do you get on Claude Pro?
Anthropic does not publish a number. Pro has a session limit that resets every five hours and a weekly limit across all models, and how far they stretch depends on conversation length, attachments and the model you pick. Claude Code draws from the same limits.
Is Claude cheaper than ChatGPT?
The entry paid plans match at $20 a month. Claude Max at $100 for 5x lines up with ChatGPT Pro's $100 tier, and ChatGPT's $200 tier is currently paused for new sign ups. On the API side Claude Sonnet 5 at $2 and $10 per million tokens is in the same band as OpenAI's mid tier models. Our ChatGPT Plus vs Pro post has the OpenAI side in detail.
Can I use Claude models on Krater?
Yes. Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5 and the Claude 4.x models are in Krater's catalogue and billed in credits per answer. On 23 September 2026 a Sonnet 5 answer cost between 0.6 and 3 credits in our run, out of 1,500 credits on the $20 Pro plan.
Related reads
- Claude usage limits explained: message, weekly and five hour caps
- ChatGPT Plus vs Pro: what the $80 gap buys, tested
- Claude vs Gemini vs ChatGPT: which model to choose
- Claude Fable 5.1 on Krater
- Krater pricing: Pro, Ultra and Max
Bottom Line
Claude Pro is $20 a month, Max is $100 or $200, and the API is a separate per token bill starting at $1 per million input tokens. The limits are relative and shared with Claude Code, so buy the plan for how often you hit the wall, not for a message count nobody publishes. If Claude is your only model, subscribe to Anthropic directly. If you want Claude alongside everything else, Krater runs the same models on one credit balance from $20.