Claude Pricing: Every Plan and API Rate Explained

Claude pricing splits into two systems: consumer chat plans and API token rates. Chat plans run from Free to Enterprise. API rates are billed per million tokens and vary by model, from Haiku 4.5 at the low end to Fable 5 at the top.

The plan you need depends on how you use Claude. Casual chat users are fine on Free or Pro. Teams running Claude Code or heavy agent workflows usually need Max or a Team seat with API access layered on top.

This guide covers every current tier, the exact API price per model, and which tier unlocks which model, so you are not guessing before you commit to a subscription.

What are the Claude pricing plans?

Claude has five consumer and business plans: Free, Pro, Max, Team, and Enterprise. Each one raises the usage ceiling and unlocks more models on top of the last.

Free costs nothing and gives you Claude Sonnet 5 with a capped weekly usage limit. Pro starts at $17 a month billed annually, or $20 a month billed monthly, and adds Claude Opus 5 access along with Claude Code. Max starts at $100 a month and multiplies your usage allowance 5x or 20x over Pro, with Opus 5 set as the default model.

Team and Enterprise are built for organizations. Team charges per seat, with standard and premium seat types you can mix. Enterprise adds SSO, HIPAA-ready configurations, and compliance controls on top of Team, billed at $20 per seat plus API-rate usage.

• Free: $0/month, Sonnet 5 access, capped weekly limits
• Pro: $17/month (annual) or $20/month (monthly), adds Opus 5 and Claude Code
• Max 5x / Max 20x: from $100/month, Opus 5 as default, 5x or 20x the Pro usage cap
• Team: $20-25/month per standard seat, $100-125 for premium seats
• Enterprise: $20/seat plus usage billed at API rates, SCIM and compliance features

What is the difference between Claude Free and Pro?

Claude Free gives you Sonnet 5 with a small weekly usage cap, while Pro raises that cap sharply and unlocks Opus 5. Free users get roughly half the usage allowance Pro gets on the same models.

Pro also adds Claude Code, unlimited projects, and access to Research and Claude for Microsoft 365. If you hit the Free tier limit mid-conversation more than once a week, Pro pays for itself in avoided friction.

Neither Free nor Pro includes Claude Fable 5. That model sits behind Max, Team, or the API.

• Free: Sonnet 5 only, 200k context window, roughly 50% of Pro usage limits
• Pro: adds Opus 5, Claude Code, unlimited projects, Research access
• Neither tier includes Fable 5

What do Claude Max 5x and Max 20x include?

Claude Max is the highest consumer tier, starting at $100 a month for 5x the Pro usage allowance, with a 20x option for heavier users at the same starting price point tiered up. Max sets Opus 5 as the default model instead of Sonnet 5.

Max also gives you higher output limits per response and priority access during high-traffic periods, which matters if you run long agentic sessions in Claude Code and keep hitting rate limits on Pro.

Most solo developers do not need Max. Teams running multiple parallel Claude Code sessions, or anyone doing long-context research work daily, are the ones who actually exhaust the Pro allowance.

• Max 5x: from $100/month, 5x Pro usage allowance
• Max 20x: from $100/month, 20x Pro usage allowance
• Opus 5 is the default model on Max instead of Sonnet 5
• Higher output limits and priority access during peak traffic

How much does Claude Team and Enterprise cost?

Claude Team charges per seat: standard seats run $20/month billed annually or $25/month billed monthly, and premium seats run $100/month annually or $125/month monthly. Teams can mix standard and premium seats across the same workspace.

Enterprise starts at $20 per seat plus usage billed at API rates, and layers on SCIM provisioning, HIPAA-ready configurations, a compliance API, and IP allowlisting on top of everything Team includes.

The seat-plus-usage structure on Enterprise means your real bill scales with how much your team actually runs through the API, not a flat per-seat number. Budget for usage separately from seats.

• Team standard seat: $20/month (annual) or $25/month (monthly)
• Team premium seat: $100/month (annual) or $125/month (monthly)
• Enterprise: $20/seat plus API-rate usage
• Enterprise adds SCIM, HIPAA-ready setup, compliance API, IP allowlisting

What is Claude API pricing per model?

Claude API pricing runs from $1 per million input tokens on Haiku 4.5 up to $10 per million input tokens on Fable 5, with output tokens costing 5x the input rate across every model in the family.

Claude Sonnet 5 is running introductory pricing of $2 input / $10 output per million tokens through August 31, 2026, after which it reverts to its standard $3 input / $15 output rate. Anyone budgeting past that date should plan for the higher number.

Claude Mythos 5 shares Fable 5's pricing and specs but is not self-serve. It ships only through invitation-only access inside Anthropic's Project Glasswing program for defensive cybersecurity workflows, so most builders will never see a Mythos 5 invoice.

• Haiku 4.5: $1 input / $5 output per million tokens, 200k context
• Sonnet 5: $2 input / $10 output through Aug 31, 2026 (then $3 / $15), 1M context
• Opus 5: $5 input / $25 output per million tokens, 1M context
• Fable 5: $10 input / $50 output per million tokens, 1M context
• Mythos 5: same pricing as Fable 5, invitation-only via Project Glasswing

How do prompt caching and batch pricing cut Claude API costs?

Prompt caching and the Batch API both cut Claude API costs, and you can stack them on the same request. Cached input tokens bill at roughly 10% of the standard input rate, and the Batch API cuts both input and output prices by 50% for workloads that can tolerate async processing.

Caching pays off when you send the same long context repeatedly, like a fixed system prompt or a reused document, since only the new tokens hit full price. Batch pricing pays off for large non-interactive jobs like classifying a backlog of support tickets overnight.

Combine both correctly and a workload that looks expensive at list price often runs at a fraction of the sticker rate. The catch is engineering effort: caching requires structuring prompts so the reused portion stays identical byte-for-byte, and batch requires your workflow to accept delayed results.

• Prompt cache hits: roughly 10% of standard input token price
• Batch API: 50% off both input and output token prices
• Caching and batch discounts can be combined on the same workload

Which Claude model should you use for which task?

Use Haiku 4.5 for high-volume, low-complexity tasks, Sonnet 5 for most day-to-day coding and writing work, and Opus 5 or Fable 5 for the hardest agentic and research problems. Model choice affects your bill more than plan choice does for API users.

Sonnet 5 is Anthropic's stated balance of speed and intelligence, and at its current introductory rate it is the cheapest way to get near-frontier output. Opus 5 is built for complex agentic coding and enterprise workloads where accuracy on multi-step tasks matters more than cost per token.

When we route tasks across the routine automation portfolio we run for client sites, the pattern holds outside the lab too: the cheapest model that clears the accuracy bar for a given task wins on cost per correct output, not the cheapest model per token. A $1/$5 model that needs three retries can cost more than a $3/$15 model that gets it right the first time.

• Haiku 4.5: high-volume, latency-sensitive, simpler tasks
• Sonnet 5: default choice for coding, writing, and most agent work
• Opus 5: complex agentic coding, enterprise workloads, deep reasoning
• Fable 5: the hardest long-running agent and research problems

What does a real Claude API workload cost?

A typical coding-agent session processing 500k input tokens and generating 50k output tokens on Sonnet 5 costs roughly $1.75 at introductory pricing, versus about $4 on Opus 5 for the same token counts. Multiply that across dozens of sessions a day and the model choice compounds fast.

The bigger cost driver for most teams is not the per-token rate but retry volume: a model that gets a multi-step task wrong on the first pass and needs two more passes to fix triples your effective spend on that task, regardless of which model you started with.

Track cost per successfully completed task, not cost per token, when comparing models across a real workload. That is the number that actually predicts your monthly bill.

• 500k input + 50k output on Sonnet 5 (intro pricing): about $1.75
• 500k input + 50k output on Opus 5: about $4.00
• Retries on a failed multi-step task multiply effective cost regardless of base rate

Can you switch between Claude plans or use both chat and API?

Yes, Claude plans and API billing are separate systems you can run in parallel. A Pro or Max subscription covers chat.claude.ai and Claude Code usage under its own allowance, while API usage through the Console bills separately per token regardless of your chat plan.

Many teams run a Team or Enterprise seat for chat access and a separate API key for production integrations, since the two systems do not share a usage pool. Switching consumer plans (Free to Pro to Max) takes effect immediately and prorates the difference.

If your team is deciding between paying for more seats versus building direct API integrations, the seat model is cheaper for interactive chat use and the API is cheaper per unit for programmatic, high-volume tasks.

• Chat plans (Free/Pro/Max/Team) and API billing are separate, non-pooled systems
• Plan upgrades and downgrades take effect immediately with prorated billing
• Programmatic/high-volume use is almost always cheaper on the API than on a chat seat