Claude Pricing in 2026: Every Plan and API Rate, Verified
Most people should buy Pro at $20 a month — it includes Claude Code and the same models as every higher tier. Max at $100–200 only pays off when session limits actually interrupt your work, and if you're building a product, the API runs from $1 per million input tokens on Haiku to $10 on Fable 5.
Verified 2026-08-13 · official pricing page · Claude in the directory
Quick pricing verdict
- Starting price
- $0 free plan; $20/mo for Pro
- Free plan
- Yes, for the apps. The API has trial credits but no permanent free tier
- Best plan for most people
- Pro — $20/mo, or $17/mo billed annually
- Plan to avoid first
- Max 20x at $200 before you've metered a month on Max 5x
- Biggest hidden cost
- Web search at $10 per 1,000 searches, charged on top of tokens
- Second hidden cost
- Claude 4.7 and later use a tokenizer that produces ~30% more tokens for the same text
- Cheapest real path
- API on Haiku 4.5 with prompt caching — $1/MTok input, $0.10 on a cache hit
- Pricing checked
- 2026-08-13, against Anthropic's own pages
Anthropic sells the same models three different ways, and the cheapest one depends entirely on how you work rather than on which model you want. A subscription bundles usage into a flat monthly fee. The API meters every token. And the gap between a well-configured API call and a careless one is larger than the gap between the cheapest and most expensive model.
This page is the reference: every current plan, every API rate, the charges that never appear on the pricing page, and worked scenarios with the arithmetic printed so you can check them. We use these plans ourselves to build this site, so the recommendations are the ones we act on.
Plans compared
| Plan | Price | Best for | Key limits |
|---|---|---|---|
| Free | $0 | Trying Claude on web, desktop, iOS, and Android | Lower usage limits than any paid tier |
| Pro | $20/mo$17/mo billed annually | Most individuals — includes Claude Code, Cowork, Design, Science | More usage than Free; unlimited projects and Research |
| Max 5x | from $100/mo | Daily agent coding that keeps hitting Pro's ceiling | 5× Pro's usage per session, higher output limits, early access |
| Max 20x | $200/mo | Near-continuous agent work | 20× Pro's usage per session |
| Team | $25/seat/mo$20/seat/mo billed annually | Small teams needing SSO and admin controls | More usage than Pro; premium seat option at $100/mo |
| Enterprise | Custom | Organisations needing SCIM, audit logs, data controls | Seat price plus usage at API rates; HIPAA-ready option |
Source: Claude pricing page, checked 2026-08-13. See the archived page we read.
What each plan includes
Free
- Includes
- Claude on web, desktop, iOS, and Android at the lowest usage tier.
- Missing
- Claude Code, the higher output limits, and any meaningful agent headroom.
- Avoid if
- You code with Claude more than occasionally — you'll hit the ceiling the same day.
- Verdict
- A trial, not a plan. Useful to confirm you like the model before paying.
Pro — $20/mo
- Includes
- Everything in Free plus substantially more usage, Claude Code, Cowork, Design, Science, unlimited projects and Research.
- Missing
- The 5× and 20× usage multipliers, and early access to new features.
- Avoid if
- Five-hour session limits interrupt real work several times a week.
- Verdict
- The default best buy. Most people never legitimately outgrow it.
Max 5x — from $100/mo
- Includes
- Five times Pro's per-session usage, higher output limits, early feature access.
- Missing
- The 20× headroom, which only continuous agent work needs.
- Avoid if
- You haven't yet measured how often Pro actually stops you.
- Verdict
- The right first upgrade for daily Claude Code users — start here rather than at $200.
Max 20x — $200/mo
- Includes
- Twenty times Pro's per-session usage.
- Missing
- Nothing above it short of Enterprise; the models are identical to Pro's.
- Avoid if
- Max 5x still leaves headroom at the end of your week.
- Verdict
- Worth it only when the $100 tier demonstrably interrupts you. Meter first.
Team — $25/seat/mo
- Includes
- All Claude features with more usage than Pro, plus SSO, admin controls, and enterprise search. A premium seat is $100/mo.
- Missing
- SCIM, audit logs, and custom data-retention controls — those are Enterprise.
- Avoid if
- You're two people who could each buy Pro and skip the admin overhead.
- Verdict
- Buy it for the controls, not the usage. Below roughly five seats, Pro seats are cheaper.
API rates by model
Prices per million tokens (MTok). The cache columns are what make an agent workload affordable: a cache hit costs 10% of the standard input price.
| Model | Input | Cache write (5m / 1h) | Cache read | Output |
|---|---|---|---|---|
| Claude Fable 5 | $10 | $12.50 / $20 | $1 | $50 |
| Claude Opus 5 | $5 | $6.25 / $10 | $0.50 | $25 |
| Claude Opus 4.8 | $5 | $6.25 / $10 | $0.50 | $25 |
| Claude Sonnet 5 | $2 | $2.50 / $4 | $0.20 | $10 |
| Claude Sonnet 4.6 | $3 | $3.75 / $6 | $0.30 | $15 |
| Claude Haiku 4.5 | $1 | $1.25 / $2 | $0.10 | $5 |
The Batch API takes 50% off both input and output, and stacks with caching.
Source: Anthropic platform pricing docs, checked 2026-08-13. See the archived page we read.
Add-ons, overages, and surprise costs
| Cost item | Trigger | Published amount | Billing behaviour |
|---|---|---|---|
| Web search | Using the server-side web search tool | $10 per 1,000 searches | Charged on top of tokens; results count as input tokens in later turns |
| Code execution | Running the tool without web search or web fetch | 1,550 free hours/org/month, then $0.05 per container-hour | Billed by execution time with a 5-minute minimum, even if the tool isn't called |
| Managed Agents runtime | Running a Claude Managed Agents session | $0.08 per session-hour | Accrues only while the session status is running, not while idle |
| Fast mode | Setting speed to fast on Opus 5 or Opus 4.8 | $10 input / $50 output per MTok | Double the standard rate; stacks with caching and residency multipliers |
| US data residency | Pinning inference_geo to us on Claude 4.6+ | 1.1× multiplier | Applies to every token category — input, output, cache writes and reads |
| Tokenizer change | Using Claude 4.7 or later | ~30% more tokens for the same text | Not a fee, but it inflates every per-token comparison against older models and other vendors |
None of these appear on the consumer pricing page; all are published in Anthropic's platform pricing docs.
Source: Platform pricing docs, checked 2026-08-13. See the archived page we read.
Real cost scenarios
| Scenario | Arithmetic | Cost |
|---|---|---|
| 10,000 support conversations on Haiku 4.5 (~3,700 tokens each) | Anthropic's own published worked example | ~$37 |
| One-hour Managed Agents session on Opus 5, 50K in / 15K out | (50K × $5 + 15K × $25) ÷ 1M + 1 hr × $0.08 | $0.705 |
| Same session with 40K of the input served from cache | (10K × $5 + 40K × $0.50 + 15K × $25) ÷ 1M + $0.08 | $0.525 |
| Heavy agent month — 100M input (90% cached) + 20M output, Haiku 4.5 | 90M × $0.10 + 10M × $1 + 20M × $5, per MTok | $119 |
| The same month on Sonnet 5 | 90M × $0.20 + 10M × $2 + 20M × $10, per MTok | $238 |
| The same month on Fable 5 | 90M × $1 + 10M × $10 + 20M × $50, per MTok | $1,190 |
The first three are Anthropic's published examples; the last three are our arithmetic from the rate table above, shown so you can reproduce them. The spread between Haiku and Fable on identical work is roughly tenfold — before batching.
Which plan should you choose?
- You chat with Claude a few times a week
- Free — The paid tiers buy usage headroom you won't touch.
- You use Claude daily, including some coding
- Pro — $20/mo — Includes Claude Code and the same models as every tier above it.
- You run Claude Code most days and keep hitting limits
- Max 5x — from $100/mo — Five-hour lockouts cost more than the upgrade once they're weekly.
- Agents run for hours, most days
- Max 20x — $200/mo — Only after a metered month proves Max 5x isn't enough.
- You're embedding Claude in your own product
- API — Turn on prompt caching first; it's a 90% discount on repeated input.
- Small team needing SSO and admin controls
- Team — $25/seat/mo — Buy it for governance. Under about five seats, individual Pro plans cost less.
How to avoid overpaying
Max 20x as your first upgrade
Step to Max 5x, run a usage meter for two weeks, and move up only if you're genuinely hitting the ceiling. Paying $200 for headroom you never touch is the most common way to overspend on Claude.
The API for ordinary personal use
If you're working inside Anthropic's own apps, the subscription is priced well below the token value a heavy user extracts. The API wins for spiky, automated, or embedded workloads — not for chatting.
Fast mode by default
It doubles the rate on Opus 5 and Opus 4.8 for faster output. Turn it on for latency-sensitive work, not for batch jobs where nobody is waiting.
Pinning US residency "just in case"
The 1.1× multiplier applies to every token category. Use it when a compliance requirement demands it, not as a default.
Price anchors
| Product | Entry price | What it anchors |
|---|---|---|
| ChatGPT Plus | $20/mo | The direct subscription comparison |
| Cursor | $20/mo | An AI coding environment rather than a model subscription |
| Kimi K3 | Usage-based | A cheap open-weight coding model on the API |
| Gemini | Free tier available | Long-context work at a lower entry price |
Price anchors, not feature equivalents — they tell you what the market charges at the same tier, not that these products do the same job.
Frequently asked questions
- How much does Claude cost per month?
- Free is $0. Pro is $20 a month, or $17 a month billed annually, and includes Claude Code. Max starts at $100 a month for 5× Pro's usage and is $200 for 20×. Team is $25 per seat monthly ($20 annual), with a $100 premium seat. Enterprise is custom — seat price plus usage at API rates.
- What is Claude's API pricing?
- Per million tokens: Haiku 4.5 is $1 input and $5 output; Sonnet 5 is $2 and $10; Opus 5 and Opus 4.8 are $5 and $25; Fable 5 is $10 and $50. Cache reads cost 10% of the input rate, cache writes 1.25× for a five-minute cache or 2× for an hour, and the Batch API takes 50% off both directions.
- Is Claude Pro or the API cheaper?
- Pro is cheaper if you work inside Anthropic's own apps most days — the subscription is priced below what a heavy user would pay per token. The API is cheaper for spiky, automated, or embedded usage where you'd otherwise pay for a ceiling you don't touch. Third-party tools must use the API regardless, since subscription credentials are limited to Anthropic's own apps.
- Is Claude Max worth $200 a month?
- Only if the $100 tier's limits still interrupt you — Max 20x is built for near-continuous agent work. Start at Max 5x, meter your usage for a couple of weeks, and upgrade only when you're genuinely hitting the ceiling.
- Does Claude have a free tier?
- Yes for the apps: the Free plan gives you Claude on web, desktop, iOS, and Android at lower usage limits. For the API, new accounts get a small amount of trial credits, but there's no permanent free API tier — usage is billed per token from the first project.
- How do I reduce my Claude API bill?
- Three levers that stack. Turn on prompt caching — a cache hit costs 10% of the standard input price and pays for itself after one read. Use the Batch API for anything asynchronous, for 50% off input and output. And match the model to the task: Haiku at $1/$5 handles routine work that doesn't need Fable at $10/$50.
- What are Claude's usage limits?
- Usage resets on a five-hour session window, and since late 2025 a weekly cap applies across all models on top of that. Agent coding consumes both far faster than chat, because one instruction can fan out into dozens of model calls. Anthropic publishes the mechanism but not a token number, so the practical test is how often you're actually interrupted.
Pricing change log
What moved since we started tracking this page, newest first. A price that changes silently is the failure mode of every other pricing guide.
- 2026-08-13Sonnet 5's $2 / $10 introductory rate became the standard price; the increase to $3 / $15 that was scheduled for 1 September will not happen.
- 2026-08-13First published, verified against claude.com/pricing and the platform pricing docs.