Claude Pricing 2026: Every Plan, API Cost, and Hidden Fee Explained

Claude pricing in 2026 has seven core SKUs: Free at $0, Pro at $20/month ($17 with annual billing), Max at $100/month, Ultra at $200/month (called Max 20x until October 30, 2026), Team Standard at $25/seat, Team Premium at $125/seat, and Enterprise at $20/seat plus API usage. API rates run from $0.10 per million input tokens on Haiku 5.5 up to $50 per million output tokens on the Fable 5.1 flagship. The newest model, Claude Opus 5.5, landed on September 22, 2026 at $4/$20, undercutting the Opus 5 it pushed onto Anthropic’s Legacy list, and it is the one Anthropic now calls the daily driver. Our guide to whether Max is worth the upgrade covers the two tiers most people are actually choosing between.

This guide breaks down every Claude pricing dimension you’ll actually pay for: subscription tiers, API rates per million tokens, a unit explained in our guide to what AI tokens are, Claude Code, AWS Bedrock, prompt caching, and the Batch API. We pulled every number directly from claude.com/pricing, the official Claude API docs, and AWS Bedrock, then ran the math on what each plan really costs in real workflows. For how Claude stacks up against every other major AI tool plan-by-plan, see our complete AI pricing comparison.

The Key Takeaways

• Free tier is permanent, with limits that reset every five hours and every week rather than a fixed message count
• Claude Pro is $20/month ($17/month with annual billing) and includes Claude Code at no extra cost
• Claude Max is $100/month for 5x Pro usage, and Claude Ultra, renamed from Max 20x on October 30, 2026, is $200/month for 10x Pro weekly usage with no 5-hour limit on verified accounts
• API pricing: Fable 5.1 flagship at $10/$50 per million tokens with $0.25 cache reads, Opus 5.5 at $4/$20, Sonnet 5.5 at $2/$10 (the same rate as Sonnet 5), Haiku 5.5 at $0.10/$0.50 for prompts up to 100,000 tokens
• Prompt caching cuts input cost by up to 90% and Batch API gives a flat 50% discount

Claude Plans 2026 (Free, Pro at $20, Max at $100, Ultra at $200, Team at $25, Premium at $125, Enterprise at $20 Plus Usage)

Claude has five ways to pay. There is a free tier, three individual subscriptions (Pro at $20, Max at $100, and Ultra at $200), two Team seat types ($25 Standard and $125 Premium), an Enterprise contract at $20/seat plus usage, and the pay-per-token API. Most readers want the consumer ladder, so here it is.

Pro and Max are the two plans most individuals pick. Team and Enterprise are both seat-based, but their floors differ. Team requires a minimum of 2 members and tops out at 150 seats, while Enterprise requires a minimum of 20 seats.

Claude Free at $0/month

The Claude Free plan has no expiration and includes web, iOS, and Android chat with Claude Sonnet 5.5, which Anthropic released on September 28, 2026. For a full look at what the free tier covers, see our guide on whether Claude is free. You also get web search, file uploads (up to 20 files per chat, 30 MB max), Artifacts, and basic projects. The Dashboards template is not on Free: Claude Dashboards needs Pro or higher. On the API, Claude Sonnet 5.5 is the mid-tier model at $2/$10 per million tokens, the same rate as Sonnet 5, which Anthropic now lists as legacy.

The catch is the rolling usage window. Anthropic uses a session-based limit that resets every five hours instead of a fixed daily cap, and adds a weekly limit on top. Anthropic publishes no fixed message count for any plan, because how far a session goes depends on conversation length, the model and the size of your files. Long PDFs and image attachments burn the cap faster than short text turns.

Are you a student? Anthropic does not offer a public student discount. The realistic budget routes are the Free plan above, the pay-as-you-go API for light use, or annual billing on Pro at $17/month. Our Claude student discount guide covers every real path and the fake codes to avoid.

Claude Pro at $20/month

Claude Pro is $20/month billed monthly, or $17/month equivalent ($200 upfront) on the annual plan. That is a 15% discount for paying yearly. Annual billing is not the only lever; here are AI deals and student discounts that stack on top of it. One caveat on both figures: they are Anthropic’s own prices. Subscribe inside the Claude iOS app and Apple bills you at a higher in-app rate, which is one reason to check how to cancel an AI subscription on a Mac or iPhone before you resubscribe.

Pro adds everything the free tier has, plus:

• Claude Code terminal coding agent at no extra cost
• Claude Cowork desktop AI agent
• Claude Design for slides, prototypes, and mockups
• Projects with persistent context
• Research mode for deeper, slower answers
• Excel, PowerPoint, and Word extensions
• At least 5x the Free plan’s usage per 5-hour session, plus a weekly limit
• Access to the Opus family, Opus 5.5 included (within your usage limits)

Pro is the right pick for most knowledge workers, writers, and developers who want Claude Code without paying API rates.

Claude Max at $100/month

Max is $100/month, billed monthly only, with no annual discount. It was called Max 5x until October 30, 2026, and the allowance did not change: 5x Pro usage every 5 hours and every week. Anthropic does not turn that into a message count, so the 225-message figure that circulates is not an official number. Max adds Fable up to half your weekly limit, $100 a month in Claude Platform API credits, early access to new features and priority access at high traffic times. Max is the sweet spot for engineers who live in Claude Code or run long Claude Cowork on Pro/Max sessions but rarely hit the 5-hour window.

Claude Ultra at $200/month (Formerly Max 20x)

Ultra is $200/month, again monthly only. It is the plan Anthropic sold as Max 20x until October 30, 2026, and existing Max 20x subscribers moved to Ultra automatically with no price change. Anthropic now describes the allowance as 10x Pro usage every week with no 5-hour limit for verified accounts, where Max 20x was sold as 20x Pro. Anthropic has not explained the move from 20x to 10x, so plan around the weekly limit, which applies across all models, plus a separate weekly limit for Fable. Anthropic verifies Ultra accounts automatically from account and payment history, and an unverified account runs under a temporary limit until then. Ultra keeps everything in Max, and it makes sense if you run heavy agentic workflows in Claude Code or all-day Cowork loops that keep hitting Max’s 5-hour window. From the week of October 7, 2026, Anthropic is also rolling out a monthly API credit for the Claude Platform: $100 on Max, $200 on Ultra, and up to $500 pooled across a Team plan, usable on any Claude model.

Claude Team Standard at $25/seat/month

Team Standard is $25/seat/month monthly or $20/seat/month on annual billing, with a 2-member minimum and a 150-seat ceiling. SSO, centralized billing, admin controls, enterprise search, and a model-training opt-out are all included on this tier. Team and Enterprise are also the only tiers that can switch on Claude Tag, the shared @Claude inside Slack channels, which carries no seat price of its own and bills against the organization’s usage balance.

Claude Team Premium at $125/seat/month

Team Premium is $125/seat/month monthly or $100/seat/month annual. A Premium seat gets 5x the usage of a Standard seat and Fable up to 50% of weekly limits. Claude Code comes with both seat types, because Anthropic includes it in every paid plan, so Premium buys capacity rather than access. You can mix Standard and Premium seats inside one workspace, so engineering subteams get Premium while everyone else stays on Standard.

Claude Enterprise at $20/seat plus API usage

Enterprise comes two ways. You can buy it self-serve online without talking to sales, or go sales-assisted for tailored terms. Either route is $20/seat plus usage at API rates, billed annually, so total cost scales with actual model traffic. Anthropic’s own framing is that the seat fee covers access only and all usage is billed separately. You get everything in Team plus SCIM, audit logs, the compliance API, HIPAA readiness, IP allowlisting, and role-based access.

Claude API Pricing per Million Tokens

The Claude API is priced per million tokens (MTok) with separate input and output rates. Here are the current rates from platform.claude.com/docs/en/about-claude/pricing.

All values are USD per million tokens. Output is always 5x more expensive than input, which is the single most important rule for budgeting Claude API spend. Anthropic splits this rate card in two. Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 are the four it sells as current, and every row tagged legacy still answers on the API at the price shown, it is simply not the one the pricing page points you at. Anthropic’s API also now offers the new Claude Dreaming memory feature, currently API-only.

Two models dropped off this table in 2026. Claude Haiku 3.5 retired on 19 February 2026 and Claude Opus 4.1 hit its retirement date on 5 August 2026, after which requests to it fail on the Claude API. Anthropic points Opus 4.1 traffic at Opus 4.8, which costs $5/$25 instead of $15/$75, and Haiku 3.5 traffic at Haiku 4.5 at $1/$5. Amazon Bedrock and Google Cloud set their own retirement schedules, so Haiku 3.5 still answers on those partner clouds even though it is gone from the Claude API.

Claude Fable 5.1 and Mythos 5.1, the Flagship Tier

Claude Fable 5.1, launched September 1, 2026, is now Anthropic’s flagship model and the most expensive to run at $10/MTok input and $50/MTok output, two and a half times the Opus 5.5 rate. Flagship here is Anthropic’s own pricing position rather than a benchmark verdict, and the cheaper Opus 5.5 measures higher on Artificial Analysis, as the next section shows. It holds the list price Fable 5 charged, and the only line that moved is the cache read, down 75% to $0.25. The Batch API cuts the base rate to $5/$25, so heavy pipelines can soften the premium. Our full Claude Fable 5.1 and Mythos 5.1 breakdown covers the benchmarks and the safeguard split. On the subscription side, Max, Ultra and Team premium seats include Fable 5.1 up to 50% of weekly usage limits, while Pro and Team standard seats reach it through pay-as-you-go usage credits rather than flat-rate access.

Claude Mythos 5.1 sits alongside Fable 5.1 at the same $10/$50 pricing, but it ships as invitation-only, trusted-access. It is not on the open API for general accounts, so treat its rate card as a reference point rather than a model you can switch to today. One exception is worth knowing: the same unrestricted configuration powers Claude Security, which any Claude Enterprise customer can buy without a Mythos invitation.

Claude Opus 5.5 arrived on September 22, 2026 at $4/MTok input and $20/MTok output, cheaper than the Opus 5 it replaced and cheaper than any Opus before it. Anthropic moved Opus 5 onto its Legacy list the same day; it still answers at $5/$25, it is simply no longer the model the pricing page points you at. The gap is not only price. On the Artificial Analysis Intelligence Index v4.3.2, Opus 5.5 scores 57.6 at max effort against 50.8 for Opus 5, and it also clears Fable 5.1 at 53.4 while costing 60% less. Batch drops Opus 5.5 to $2/$10, and cache hits land at $0.20, a 5% multiplier instead of the usual 10%. If you are picking one Claude model for production work today, this is the price-performance pick; see our full Claude Opus 5.5 breakdown for the benchmark detail.

Important Caveat on the New Tokenizer

Opus 4.7 and later Opus models, every Fable and Mythos model, Sonnet 5, and Haiku 5.5 all ship with a new tokenizer that produces roughly 30% more tokens for the same fixed text. The per-million rate looks unchanged on paper, but your real bill per request goes up because each request now counts more tokens. Claude Opus 4.7 introduced it and everything newer inherits it; Sonnet 4.6 and earlier keep the old tokenizer. Re-estimate your token math before swapping to any of the newer models.

Long Context and Fast Mode

The full 1 million token context window on Opus 5.5, Opus 5, Fable 5.1, Mythos 5.1, Fable 5, Mythos 5, Opus 4.8, Opus 4.7, Opus 4.6, Sonnet 5.5, Sonnet 5, and Sonnet 4.6 is included at standard pricing. Anthropic’s own wording is that Claude 4.6 and later models include the full 1M window at standard rates. A 900K-token request is billed at the same per-token rate as a 9K request, so there is no long-context premium on the Claude API for these models. Haiku 5.5 is the exception: it has a 1M window too, but a prompt over 100,000 tokens pays five times the rate, $0.50/$2.50 instead of $0.10/$0.50.

Fast mode bills at 2x standard rates across the full context window, so $8/MTok input and $40/MTok output on Opus 5.5 and $10/$50 on Opus 5 and Opus 4.8. Anthropic runs it as a research preview on the first-party Claude API only, so it is not available on Claude Platform on AWS, Amazon Bedrock, or Google Cloud, and it is not offered on Opus 4.7 or Opus 4.6. Fast mode also cannot be combined with the Batch API, though prompt caching multipliers still stack on top of it. It is worth the premium only when latency matters more than cost.

How to Cut Your Claude API Bill: Caching and Batch

Two features cut your token spend by 50-90%.

Prompt Caching

Cache reads cost just 10% of the standard input price on most models, with two exceptions. Fable 5.1 and Mythos 5.1 charge $0.25, or 2.5% of input, and Opus 5.5 and Sonnet 5.5 charge 5%, which is $0.20 and $0.10. Anthropic halved the Sonnet 5.5 cache read from $0.20 on October 7, 2026. Cache writes cost 1.25x base input (5-minute TTL) or 2x (1-hour TTL). For workloads with large repeated context, like agent loops, RAG over the same docs, or long system prompts, caching pays off after a single cache hit on the 5-minute tier.

Worked example: 50,000 input tokens on Opus 5.5 with 40,000 of them cached:

• Uncached: 50,000 × $4 / 1M = $0.20
• Cached: 10,000 × $4 + 40,000 × $0.20, all per million = $0.048

That’s a 76% drop on input cost for the same request. Output tokens are unaffected.

Batch API

The Batch API processes requests asynchronously inside a 24-hour window in exchange for a flat 50% discount on both input and output. It works on every Claude model, every tier.

Batch is ideal for overnight document processing, dataset labeling, and any workload where same-second responses don’t matter.

How Much Does Claude Code Cost?

Claude Code is not sold as a separate subscription. It rides on top of an existing Claude account in one of four ways:

• Pro plan at $20/month (Code included)
• Max at $100/month or Ultra at $200/month (Code included with higher usage, and no 5-hour limit on verified Ultra accounts)
• Team at $20/seat/month Standard or $100/seat/month Premium on annual billing (Code included on both, Premium with 5x the usage)
• Pay per token via the Claude API at standard model rates with no monthly floor

There is no free Claude Code plan and no public Claude Pro free trial; see our AI free trials master list for what does have one. If you only run Code occasionally and don’t need consumer chat, the API path is cheaper. If you use it daily for serious agentic work, Max is the sweet spot for most engineers because Code burns through Pro limits quickly.

The Pro/Max plan path has fixed monthly cost, predictable budget, and no token math. The API path has no upper bound but works out cheaper for low-volume use. Pick based on usage pattern, not gut feel.

AWS Bedrock Claude Pricing

Anthropic Claude is available on AWS Bedrock, Google Vertex AI, and Microsoft Foundry. Per-token rates on Bedrock match the direct Claude API rates for most models, with two important wrinkles.

Wrinkle 1, regional surcharges. Starting with Sonnet 4.5 and Haiku 4.5, Bedrock offers two endpoint types. Global endpoints are the default and route dynamically; regional endpoints guarantee data routing through a specific region and add a 10% premium over global pricing. Earlier models retain their existing pricing regardless of endpoint.

Wrinkle 2, cross-region inference. Enabling cross-region inference adds another routing surcharge. AWS publishes the live numbers on the Bedrock pricing page; check there before estimating spend.

Bedrock supports on-demand (pay per token), batch (50% off, same as direct API), and provisioned throughput (reserved capacity) billing modes. The 90% cache discount also carries over.

When to pick Bedrock over the direct Claude API:

• You already run on AWS and want consolidated billing
• You need data residency in a specific region
• You want IAM-based access control without custom auth

When to stay on the direct Claude API:

• You want the cheapest possible global endpoint
• You need the latest models on day one (Bedrock often lags by weeks)
• You’re using features like Claude Managed Agents that ship API-only

Claude Pricing vs ChatGPT

Both Claude Pro and ChatGPT Plus cost $20/month at the standard tier, so the headline price is identical.

Two practical differences. First, both $20 tiers now ship a terminal coding agent, Claude Code on Pro and Codex on Plus, so the question is usage limits and model access rather than whether you get one at all. Second, Claude is ad-free across every tier, while ChatGPT now runs ads on Free and Go. If you want Claude Code specifically, Pro is the only $20 route to it. If you mostly use multimodal image generation, ChatGPT wins on default toolset. For the full ChatGPT plan ladder including the $100, $200 and $500 Pro tiers and the $8 Go plan, see our ChatGPT pricing guide.

Enterprise Add-ons and Hidden Costs

A few line items that often catch buyers by surprise:

• Web search tool on the API: $10 per 1,000 searches plus standard token cost for the returned content
• Code execution tool: free with web search/web fetch; standalone billed at $0.05/hour per container beyond 1,550 free hours/month per organization
• Claude Managed Agents session runtime: $0.08 per session-hour on top of token costs
• US-only data residency on the Claude API: 1.1x multiplier on all token categories
• Tool use system prompt overhead: roughly 290-475 tokens added per request when tools are enabled, depending on model and tool choice

These look small per call but stack on agentic workloads that run thousands of tool calls.

A Cheaper Alternative: One Subscription, All the Models

Here’s the honest tradeoff with Claude pricing. If you want Claude plus other models, like ChatGPT, Gemini, Grok, and DeepSeek, you’d stack Claude Pro $20, ChatGPT Plus $20, Google AI Pro $19.99, plus Grok and DeepSeek paid tiers. That’s roughly $80-$100/month just to keep accounts open. If you are choosing between just two of them, see whether Gemini is the better value.

Fello AI gets you Claude, ChatGPT, Gemini, Grok, and DeepSeek on one bill of $11.99/month, in one Mac, iPhone, and iPad app. It is featured in our best AI apps for iPhone ranking. You keep one prompt and switch the model under it, with no juggling logins. It’s not a replacement for Claude Pro if you live in Claude Code, but for everyday chat and side-by-side model comparisons, the math is hard to beat.

Conclusion

The simplest version of Claude pricing in 2026: $0 Free, $20 Pro, $100 Max, $200 Ultra, per-seat Team and Enterprise, plus per-token API rates that scale from $0.10/MTok input on Haiku 5.5 to $50/MTok output on the Fable 5.1 flagship. Pick Pro for everyday use, Max for heavy Code/Cowork workloads, Ultra when the 5-hour window is what stops you, and the API only when you want pay-as-you-go.

If you want one bill that covers Claude and every other major model, try Fello AI at $11.99/month. Otherwise, head to claude.com/pricing, pick the tier that matches your usage window, and start building. If you are on a Mac, our Claude desktop client guide for macOS covers running Claude natively once your plan is sorted.

FAQ