Claude Pricing in 2026: Every Plan, API Rate and the September Increase

Key Takeaways

• A subscription is only fixed until someone hits a limit. Enterprise seats cover access only, and all usage bills at API rates on top.

• Chat, Claude Code and Cowork draw from one shared pool, so a few heavy Claude Code users can quietly consume the allowance an entire team depends on.

• Batch processing halves the cost and prompt caching cuts repeat input to a tenth. Most teams pay list price because nobody owns the optimisation.

• Sonnet 5 is on introductory pricing until 31 August 2026. On 1 September it goes from $2/$10 to $3/$15 per million tokens, a 50% rise on the model most teams default to.

• Guickly shows every Claude seat, token and dollar next to the rest of your AI stack, and connects each dollar to what it produced.

Claude's price list looks simple. Four subscription tiers and a table of API rates. The bill is where it stops being simple, because two things move underneath you: which model your team reaches for, and whether anyone notices when a rate changes.

One is changing in four weeks.

Note: Prices checked on Aug 02, 2026, against Anthropic's official pricing page.

How much does Claude cost per month?

Two details that catch people out.

Max is sold as a multiplier, not an amount. Anthropic describes the tiers as 5x and 20x Pro capacity per session. There is no published token or message figure behind those numbers, so you cannot forecast from them. You find your ceiling by hitting it.

Enterprise seats do not include usage. In Anthropic's own words, the seat fee covers access to the platform and does not include any usage. Headcount sets the floor of an Enterprise bill. Usage sets the ceiling. Two companies with identical seat counts can pay very differently.

What are Claude's API rates in 2026?

Per million tokens.

The spread is the point. Fable 5 output costs ten times Haiku 4.5 output. Most production traffic does not need the top of that table, and most of it goes there anyway because the default was set once and never revisited.

What is changing on 1 September 2026?

Sonnet 5's introductory pricing ends. Anthropic's wording: introductory pricing of $2/$10 per million input/output tokens is in effect through 31 August 2026, after which the standard pricing of $3/$15 will take effect.

If Sonnet 5 is your workhorse, and for most teams it is, budget for a 50% increase on that line from September. Nothing about your usage has to change for your bill to rise by half.

This is the pattern worth internalising. Model pricing now moves on a scale of weeks. If the only way you find out is the invoice, you find out a month late.

What else is billable?

The line items nobody models, all first-party:

Read that table as an optimisation list rather than a cost list. Cache reads at a tenth of input price and batch at half are the two largest levers available, and both are configuration rather than engineering.

Claude vs ChatGPT: which is cheaper?

Wrong question, and it is the question that produces bad decisions.

On comparable tiers the headline subscription prices sit close enough that the difference is noise. What actually moves your bill is which model handles which workload, whether caching is on, and whether batch is used for anything that is not interactive. A team on the more expensive provider with routing and caching in place will usually pay less than a team on the cheaper one that sends everything to the frontier model.

Pick on capability for your workload. Then manage the routing, because that is where the money is.

Why does a fixed subscription produce a variable bill?

Because the subscription is a floor with a shared pool sitting on top.

Chat, Claude Code and Cowork draw from the same allowance. A handful of engineers running long Claude Code sessions can consume what the rest of the team was relying on, and nobody sees it happen until people start hitting limits. On Enterprise it is more direct: seats buy access, usage bills separately at API rates.

Which means the useful question is not what Claude costs. It is what Claude costs you, by team, by model, next month.

Most companies cannot answer that, because Claude is one line in a stack that also holds OpenAI, Copilot, Gemini and a dozen tools somebody bought on a card. Guickly answers it in one view, and connects each dollar to what it actually produced. If a 50% rise on your main model would be a surprise rather than a plan, that is the gap.

FAQ

How much does Claude cost per month? Free is $0. Pro is $20 a month or $200 a year. Max is $100 a month for the 5x tier and $200 for 20x. Team is $25 per seat monthly or $20 on annual for standard seats, and $125 monthly or $100 annual for premium seats that include Claude Code. Enterprise is $20 per seat per month billed annually, plus all usage at API rates.

What are Claude's API prices in 2026? Per million input and output tokens: Fable 5 $10 and $50, Opus 5 $5 and $25, Sonnet 5 $2 and $10 until 31 August 2026 then $3 and $15, Haiku 4.5 $1 and $5. Opus 4.5 through 4.8 match Opus 5 at $5 and $25.

Is Claude Sonnet 5 getting more expensive? Yes. Sonnet 5 is on introductory pricing of $2/$10 per million tokens through 31 August 2026. From 1 September 2026 standard pricing of $3/$15 applies, a 50% increase. Usage does not need to change for the bill to rise.

Does the Claude Enterprise seat price include usage? No. Anthropic states the seat fee covers platform access only and does not include usage. Usage bills separately at API rates, so headcount sets the floor of the bill and usage sets the ceiling.

How many messages do you get on Claude Max? Anthropic publishes Max as 5x and 20x Pro capacity per session rather than an absolute number. There is no published token or message figure, so the limits cannot be forecast precisely and are usually discovered by hitting them.

How can you reduce Claude API costs? Four levers, in order of effort. Batch processing gives a 50% discount on anything not interactive. Prompt caching drops repeat input to 0.1x on reads. Model routing moves work off the frontier tier when a cheaper model handles it as well. And visibility, because you cannot optimise routing you cannot see.

Is Claude cheaper than ChatGPT? On comparable tiers the headline prices are close enough that the difference rarely decides the bill. Model routing, caching and batching move it far more. A team on the pricier provider with routing in place typically pays less than a team on the cheaper one sending everything to a frontier model.

Why is my Claude bill variable on a fixed plan? Chat, Claude Code and Cowork share one allowance, so heavy Claude Code use can consume a team's pool. On Enterprise, seats and usage are billed separately, so usage growth shows up directly on the invoice.