Claude Pricing Guide: Every Plan Compared

The average Claude Code developer spends roughly $13 per active coding day, which adds up to $150 to $250 per month before applying optimization techniques like prompt caching or model selection. That range catches teams off guard when they expected Claude to cost about as much as a streaming subscription. The gap between expectation and reality exists because Anthropic now offers six distinct tiers, each with different usage ceilings, feature sets, and billing models.

Here is the full lineup as of June 2026:

• Free: $0. Web, desktop, and mobile chat with limited daily usage. Includes code generation, data visualization, web search, conversation memory, file creation, code execution, and extended thinking.
• Pro: $20/month, or $17/month billed annually. Adds Claude Code, Cowork, Design, Research mode, unlimited projects, additional model choices including Opus, and integrations with Microsoft 365.
• Max 5x: $100/month. Five times the Pro usage ceiling per session, plus priority access during peak traffic and higher output limits.
• Max 20x: $200/month. Twenty times the Pro usage ceiling. Built for developers and researchers who use Claude as their primary tool throughout the day. Includes early access to advanced features.
• Team Standard: $25/seat/month, or $20/seat billed annually. Pro-level features with SSO, admin controls, enterprise search, central billing, and desktop app deployment.
• Team Premium: $125/seat/month, or $100/seat billed annually. 5x usage per seat plus Claude Code access. Requires a minimum of five seats.
• Enterprise Self-Serve: $20/seat plus API-rate usage costs. Adds SCIM, audit logs, compliance API, data retention controls, network access controls with IP allowlisting, and a HIPAA-eligible option.
• Enterprise Sales-Assisted: Custom pricing with MSA support and flexible commitment structures.

Anthropic also offers an Education plan with institution-wide access at discounted rates, including research mode and API credits for students, faculty, and staff.

Prices do not include applicable tax. All plans use a rolling usage window rather than a hard monthly cap, which means your available capacity resets on a rolling basis rather than on a calendar date. The sections below cover each tier in depth, including the API token pricing that developers building on Claude's platform need to budget for.

Claude's free tier is more capable than most people realize. You get the full chat interface across web, iOS, Android, and desktop, plus code generation, data visualization, web search, conversation memory, file creation, code execution, desktop extensions, Slack and Google Workspace integration, remote MCP connectors, and extended thinking. For someone who uses Claude a few times a week for drafting, quick code snippets, or research questions, the free tier handles it.

The ceiling shows up with sustained use. Free tier usage limits reset on a rolling window, and heavy users report hitting those limits by mid-afternoon. Once you're rate-limited, you wait. There is no option to pay for a single extra session.

Pro removes that bottleneck and unlocks five tools that the free tier does not include:

Claude Code turns Claude into an AI coding agent that runs in your terminal. It reads your codebase, writes and edits files, runs tests, handles git operations, and works with your existing development tools. For developers, this single feature often justifies the entire Pro subscription.

Claude Cowork creates persistent collaboration sessions where Claude works alongside you in a shared workspace. Unlike standard chat, Cowork sessions maintain context across tasks and let you delegate multi-step work.

Claude Design provides visual creation tools directly inside Claude. You can generate, edit, and iterate on designs without switching to a separate application.

Research mode lets Claude conduct deep, multi-source investigations. It searches the web, reads multiple sources, cross-references claims, and produces a cited report. The quality difference compared to a single web search is significant for anyone doing competitive analysis, market research, or technical investigations.

Additional model choices give you access to Opus and other models that are not available on the free tier. Different models have different strengths: Opus handles complex reasoning, Sonnet balances speed and capability, and Haiku is fast for simple tasks.

The annual billing option drops Pro from $20/month to $17/month, saving $36 per year. If you have been using Claude daily for a month and plan to continue, switching to annual billing is straightforward.

Pro also opens the door to MCP (Model Context Protocol) integrations that extend what Claude can do. For example, Claude Code can read and write files in shared workspaces through MCP servers like Fast.io, push code to GitHub, or interact with databases. S3 buckets, Google Drive, and Dropbox also work as storage backends, though MCP-native tools handle the agent-to-workspace handoff with less manual overhead.

Max plans exist because Pro's usage limits run out faster than heavy users expect. A developer pair-programming with Claude Code through a full workday, or a researcher running multi-step investigations that span dozens of sources, will hit Pro's ceiling well before the 5-hour rolling window resets.

The "5x" and "20x" labels are multipliers applied to Pro's per-session usage capacity. Max 5x at $100/month gives you five times the headroom per session. Max 20x at $200/month gives twenty times, along with higher output limits and early access to features that have not rolled out to lower tiers yet. Both include priority access during peak traffic, which matters during weekday business hours when free and Pro tier response times sometimes slow down.

Claude Code usage drives the bulk of Max-tier consumption. A single Claude Code session that scaffolds a new project, writes tests, and refactors existing code can burn through tokens quickly, especially with extended thinking enabled. The Finout team documented that autocompact events during long coding sessions consume 100,000 to 200,000 tokens each, and context resubmission loops can add 50,000 to 300,000 tokens per event. These spikes are invisible unless you track token usage closely.

When Max 5x makes sense:

• You use Claude Code 2 to 4 hours per day and occasionally hit Pro limits
• You run Research mode on complex topics that require multiple search rounds
• You want priority access but your daily usage is moderate
• Your team does not need organizational admin controls (Team tiers cover that)

When Max 20x makes sense:

• Claude Code is your primary development environment for 6+ hours daily
• You regularly run extended thinking on hard reasoning or architecture tasks
• You coordinate multiple Cowork sessions throughout the day
• You want the highest individual-account usage ceiling Anthropic offers

The cost math helps frame the decision. At $200/month, Max 20x works out to roughly $9 per workday for a developer who codes 22 days per month. If Claude Code replaces even one hour of manual development work per day at typical developer billing rates, the plan pays for itself several times over. Max 5x at $100/month is the safer starting point if you are still learning your usage patterns. You can upgrade mid-cycle if you consistently hit the 5x ceiling.

One nuance worth tracking: peak-hour usage between roughly 5 AM and 11 AM Pacific on weekdays reportedly consumes tokens at a 1.3x to 1.5x multiplier compared to off-peak hours. If your heaviest Claude Code sessions happen during that window, your effective usage ceiling is lower than the nominal 5x or 20x figure.

Give your Claude agents a persistent workspace

Fast.io provides 50GB of free storage with an MCP endpoint for Claude Code and Cowork sessions. No credit card, no trial, no expiration.

Sign Up Free

Individual plans work for solo developers and freelancers. Once your organization has multiple people using Claude, the Team and Enterprise tiers add admin controls, centralized billing, and compliance features that individual accounts cannot replicate.

Team Standard costs $25/seat/month, or $20/seat billed annually. Every seat gets access to Claude's full feature set with more usage than the Pro tier. The organizational layer includes SSO for centralized authentication, admin controls for managing members and permissions, enterprise search across all team conversations, central billing that consolidates costs under one invoice, and integrations with Slack, Google Workspace, and Microsoft 365. Admins can deploy the Claude desktop app organization-wide.

Team Premium costs $125/seat/month, or $100/seat billed annually, and requires a minimum of five seats. Each seat gets 5x usage capacity (matching the individual Max 5x tier) plus Claude Code access. This is the critical distinction: Team Standard seats do not include Claude Code. If your team needs Claude Code, every developer who uses it needs a Premium seat.

That detail catches some organizations by surprise. Consider a 15-person company where 5 engineers need Claude Code and 10 people need chat and research capabilities. You would assign 5 Premium seats at $100/seat/month and 10 Standard seats at $20/seat/month, totaling $700/month on annual billing. Putting everyone on Premium "just in case" would cost $1,500/month, more than double.

Enterprise Self-Serve starts at $20/seat and adds API-rate usage costs on top. This tier exists for organizations in regulated industries or those with strict security requirements. The feature additions over Team include:

• Role-based access controls with granular permission management
• SCIM (System for Cross-domain Identity Management) for automated user provisioning and deprovisioning
• Audit logs that track every action across the organization
• A compliance API for integrating Claude usage data into existing GRC workflows
• Data retention controls with configurable policies
• Network access controls with IP allowlisting
• HIPAA-eligible option for healthcare organizations handling protected health information
• Claude Security (currently in beta) for additional threat detection

Enterprise Sales-Assisted is for organizations that need custom contractual terms, MSA support, flexible commitment structures, volume discounts, or dedicated support. Pricing is determined through Anthropic's sales team.

When teams reach this scale, Claude's output, including generated code, research reports, analysis documents, and design assets, needs to live somewhere the entire organization can access. Individual chat histories do not serve that purpose. Some teams use Google Drive or SharePoint. Others use platforms built specifically for agent-to-human handoff. Fast.io's shared workspaces let Claude Code and Cowork sessions push output directly to a workspace through the MCP server, where teammates can review, comment, and continue the work. The free tier includes 50GB of storage and 5,000 AI credits per month with no credit card required, which covers most teams evaluating the workflow.

Developers building applications on Claude's API pay per token rather than per seat. Token pricing varies by model family, and Anthropic's current lineup spans four capability tiers.

Current model pricing (per million tokens):

• Fable 5: $10 input, $50 output. Anthropic's frontier model for the most demanding creative, analytical, and agentic tasks.
• Opus 4.8, 4.7, 4.6, 4.5: $5 input, $25 output. Strong reasoning and coding models suitable for most production workloads. Opus 4.7 and later use a new tokenizer that may use up to 35% more tokens for the same text compared to earlier models.
• Sonnet 4.6, 4.5: $3 input, $15 output. The price-to-performance sweet spot for production applications that need quality without the Opus price tag.
• Haiku 4.5: $1 input, $5 output. Fast and affordable for high-volume tasks like classification, extraction, routing, and customer support triage.

Three cost-reduction features change the economics significantly:

Batch API processes requests asynchronously at a flat 50% discount on both input and output tokens. If your application does not require real-time responses, such as nightly report generation, bulk document processing, or offline analysis, batch processing cuts costs in half. Opus models drop to $2.50/$12.50 per million tokens in batch mode, and Haiku 4.5 falls to $0.50/$2.50.

Prompt caching avoids reprocessing the same context on every request. A 5-minute cache write costs 1.25x the standard input price, but every subsequent cache hit costs only 0.1x, which is a 90% reduction. The cache pays for itself after a single hit. For applications that send the same system prompt, tool definitions, or document context across hundreds of requests, the savings compound quickly. A 1-hour cache option costs 2x on the initial write but supports longer-running sessions and workflows.

Fast mode trades cost for speed. Opus 4.6 and 4.7 in fast mode cost $30 input and $150 output per million tokens, a 6x premium over standard pricing. Opus 4.8 runs at $10/$50 in fast mode, matching Fable 5's standard rate. Fast mode targets interactive coding tools, real-time chat applications, and latency-sensitive agent loops where response time matters more than per-token cost.

A few additional API costs to budget for: web search costs $10 per 1,000 searches. Web fetch has no surcharge beyond the tokens the fetched content consumes (roughly 2,500 tokens per average web page). Claude Managed Agents add a session runtime fee of $0.08 per session-hour on top of standard token costs.

For agent pipelines that process documents, token costs for large files add up. A 500KB research paper consumes roughly 125,000 input tokens on a single read. Teams running pipelines that revisit the same documents across sessions often pair Claude's API with a persistent storage layer to cache processed results. S3, Google Drive, and Fast.io all work for this. Fast.io's Intelligence Mode auto-indexes uploaded files for semantic search, so agents can query document contents through the MCP server without re-ingesting the full text each time, which directly reduces the input token costs that dominate most agent workflows.