Anthropic pricing, plans and limits

Claude models via API, Claude.ai plans, Bedrock and Vertex AI.

Best for
Backend engineers building production AI features at software startups
Not for
Teams with air-gapped or strict on-premises data sovereignty requirements

Should you use Anthropic?

Anthropic sells Claude through its own API, Claude.ai plans, Amazon Bedrock and Google Cloud. The four current models span a cheap routing tier to a premium reasoning tier, and the Batch API halves the token price of any of them. Individual plans start free.

1M token context window on Sonnet 4.6 and Opus 4.6+ at standard per-token pricing, combined with 50% Batch API discount, making large-context high-volume inference more cost-accessible than any competitor at this capability level.

Safety-first Constitutional AI reputation drives 80% enterprise revenue share; multi-year compute and distribution deals with AWS, Google, Microsoft, and SpaceX create significant infrastructure and channel lock-in.

Also for

  • AI product managers at mid-market companies deploying customer-facing agents
  • Backend engineers at software startups building production AI features, agents, or developer-facing APIs
Show 3 more
  • Product teams deploying customer-facing AI assistants who need reliable high-context inference
  • DevOps and platform teams at enterprises requiring SOC 2 and HIPAA-compliant AI inference
  • Data scientists and researchers running large-context document analysis or structured extraction at scale

Also not for

  • Solo developers with under $50/month AI spend who only need simple completions

Main trade-off

No self-hosted deployment path; all inference runs on Anthropic or cloud-partner infrastructure, blocking air-gapped, strict-sovereignty, or on-premises deployments outright.

Anthropic pricing

Server plans

Anthropic server plans
PlanPrice
FreeSubscription · Claude chat on web, desktop and mobile with web search, file creation, memory and Artifacts; no API access included · US East · US West · EU West · EU Central · Asia Pacific$0/mo
ProSubscription · $20 billed monthly or $17 a month on an annual plan; more usage, Claude Code, Projects and Claude in Chrome · US East · US West · EU West · EU Central · Asia Pacific$20/mo
Team Standard seatSubscription · For teams of 2 to 150: $25 per seat billed monthly or $20 on an annual plan; central billing, SSO and admin controls · US East · US West · EU West · EU Central · Asia Pacific$25/mo
Max 5xSubscription · From $100 a month; 5x the usage of Pro, higher output limits and priority access at busy times · US East · US West · EU West · EU Central · Asia Pacific$100/mo
Team Premium seatSubscription · 5x the usage of a standard seat: $125 per seat billed monthly or $100 on an annual plan · US East · US West · EU West · EU Central · Asia Pacific$125/mo
Max 20xSubscription · 20x the usage of Pro; price shown at checkout on claude.com/pricing · US East · US West · EU West · EU Central · Asia Pacific—
EnterpriseSubscription · US$20 per seat a month billed annually, plus usage at API rates; spend limits and role-based access · US East · US West · EU West · EU Central · Asia Pacific—
Claude Fable 5.1API · $10 input / $50 output per MTok; cache reads $0.25; for demanding reasoning and long-horizon agentic work · US East · US West · EU West · EU Central · Asia Pacific—
Claude Opus 5.5API · $4 / $20 per MTok; fast mode at 2x standard pricing; for long-running agentic coding and knowledge work · US East · US West · EU West · EU Central · Asia Pacific—
Claude Sonnet 5.5API · $2 / $10 per MTok; the speed and intelligence middle of the lineup, full 1M-token context at standard rates · US East · US West · EU West · EU Central · Asia Pacific—
Claude Haiku 5.5API · $0.10 / $0.50 per MTok for prompts up to 100K tokens, $0.50 / $2.50 above; for classification, extraction and routing · US East · US West · EU West · EU Central · Asia Pacific—
Claude Haiku 4.5API · $1 / $5 per MTok; older model, still listed on the pricing page · US East · US West · EU West · EU Central · Asia Pacific—
Claude Fable 5.1 (Batch)API · Batch API at $5 / $25 per MTok (50% off) for asynchronous work · US East · US West · EU West · EU Central · Asia Pacific—
Claude Opus 5.5 (Batch)API · Batch API at $2 / $10 per MTok (50% off) for asynchronous work · US East · US West · EU West · EU Central · Asia Pacific—
Claude Sonnet 5.5 (Batch)API · Batch API at $1 / $5 per MTok (50% off) for asynchronous work · US East · US West · EU West · EU Central · Asia Pacific—
Claude Haiku 5.5 (Batch)API · Batch API at $0.05 / $0.25 per MTok (50% off) for prompts up to 100K tokens · US East · US West · EU West · EU Central · Asia Pacific—
Billing
Pay as you go · Subscription · Freemium · Free
How the pricing works, in full

Claude.ai: Free; Pro $20 a month ($17 annual); Max from $100 a month; Team $25 per standard seat monthly ($20 annual) and $125 per premium seat ($100 annual); Enterprise US$20 per seat a month billed annually plus usage at API rates. API per million input/output tokens: Fable 5.1 $10/$50, Opus 5.5 $4/$20, Sonnet 5.5 $2/$10, Haiku 5.5 $0.10/$0.50 for prompts up to 100K tokens ($0.50/$2.50 above), Haiku 4.5 $1/$5. Batch API is 50% off. Prices read from claude.com/pricing and the Claude Platform pricing docs on 2026-10-09.

Bandwidth pricing

Token-based billing only. No separate bandwidth or data transfer charges apply.

Costs that are easy to miss

  • Output tokens are priced at 5x the input token rate on all models, which triples effective spend relative to input costs when generating long responses or structured data
  • Enterprise plan charges $20/seat/month on top of all metered API token usage; a 50-person team adds $1,000/month before any token consumption

Anthropic regions and compliance

Uptime SLA
99.9%

Available regions

  • US East
  • US West
  • EU West
  • EU Central
  • Asia Pacific

Compliance certifications

  • SOC 2 Type II
  • ISO 27001
  • ISO 42001
  • HIPAA

As stated by the provider. Ask for the current report before you rely on one.

Managed services

  • Claude API
  • Claude.ai
  • Claude Code
  • Agent SDK
  • Files API
  • Skills API
  • Batch API
  • Message Batches API

Access methods

  • Web console
  • API
  • CLI
  • Dashboard

What Anthropic does

Anthropic sells four current Claude models, Haiku, Sonnet, Opus and Fable, and every one except Haiku includes a 1 million token context window at standard rates. Per-token prices run from cents to tens of dollars per million tokens, by model, and the Batch API takes half off. Individual plans start free, with paid tiers for heavy users and teams.

  • Million-token context at standard rates

    The current models other than Haiku include the full one-million-token window at normal per-token prices, with no long-context surcharge.

  • Batch API discount

    Asynchronous requests are billed at half the standard input and output price, which suits overnight jobs and bulk classification.

  • Prompt caching

    Cache reads cost a small fraction of the input price, and the saving stacks with batch pricing for repetitive agent workloads.

  • Managed Agents

    Hosted agent sessions bill standard token rates plus an hourly runtime charge while the session is active.

  • Server-side tools

    Web search is billed per thousand searches, and code execution is free when it is used together with web search or web fetch.

Strengths and limits

Strengths

  • One vendor covers high-volume routing work and hard reasoning work, with four current models priced accordingly.
  • Batch and cache-read discounts stack, so repetitive high-volume jobs can cost a fraction of list price.
  • Full million-token context is billed at standard rates on every current model except Haiku.

Limits

  • No self-hosted or on-premises option; all inference runs on Anthropic or cloud-partner infrastructure.
  • Output tokens cost several times the input rate on every current model, so long generated documents are the expensive case.
  • Haiku pricing steps up once a prompt passes the long-prompt threshold, which weakens the low-cost pitch for very long jobs.
  • The top Max tier and Enterprise prices are not published as a single number; they need checkout or a sales conversation.

Anthropic alternatives

The nearest AI infrastructure services on Hokai for the same job.

ServiceRegionsPricingComplianceChecked
Anthropic This pageUS · Europe · Asia PacificFree tier · Pay as you go · SubscriptionSOC 2 · ISO 27001 · ISO 42001 +19 Oct
Google Vertex AIGoogle Cloud's managed AI platform, now called Gemini Enterprise Agent Platform.US · Canada · South America · Africa +3Free tier · Pay as you go · Committed spendISO 27001 · SOC 2 · HIPAA26 Oct
LuminalCompiler-built inference for open models, as a serverless API or on-prem.--Pay as you go--23 Oct
Claude in Amazon BedrockClaude models run by AWS, billed and secured through your AWS account.Africa · Asia Pacific · Canada · Europe +3Pay as you go · Committed spendHIPAA · GDPR · FedRAMP10 Oct

Anthropic facts

Provider founded
2021
Headquarters
San Francisco, CA
Total funding
$132B+
Team size
1000-5000
Open source
No

What changed

  • Release Claude Opus 4.8 released with extended thinking and Fast Mode source
  • Update SpaceX compute deal adds 220,000+ NVIDIA GPUs, rate limits doubled source
  • Release SOC 2 Type II and ISO 27001 certifications achieved source

Questions about Anthropic

What does Anthropic's platform sell?

Anthropic sells Claude, a family of language models, through its own API, Claude.ai subscriptions, Amazon Bedrock and Google Cloud Vertex AI. Current models come in four sizes, and a buyer matches the size to the job: Haiku is the fast, cheap end, Fable the most capable. Older generations stay listed at their earlier prices.

What are Anthropic's pricing plans in 2026?

Anthropic lists these rates per million input and output tokens: Fable 5.1 $10 and $50, Opus 5.5 $4 and $20, Sonnet 5.5 $2 and $10, and Haiku 5.5 $0.10 and $0.50 for prompts up to 100,000 tokens ($0.50 and $2.50 above that). Haiku 4.5 is $1 and $5, and the Batch API takes 50% off both directions. Claude.ai plans run Free, Pro at $20 a month ($17 on an annual plan), Max from $100, and Team at $25 per standard seat monthly ($20 annual). Checked on Anthropic's pricing pages on 9 October 2026.

Can I rent GPUs from Anthropic?

No. Anthropic sells access to Claude models and does not rent raw GPU capacity. You send requests to the API, or use a Claude.ai plan, and the inference runs on Anthropic or cloud-partner infrastructure. Teams that need models running inside their own environment have to look at a different provider, or use Claude through Amazon Bedrock or Google Cloud Vertex AI, which bill separately under those clouds.

Which Claude plan or model fits which buyer?

A developer testing prompts can start on the API with Haiku or Sonnet and add the Batch API once volume is steady. Someone who uses Claude all day for coding chooses Pro or Max, since Claude Code is included from Pro upward. A small or mid-sized team picks the Team plan, and larger organizations talk to sales about Enterprise. Fable suits hard reasoning where quality matters more than cost, and anyone who must run models on their own hardware should look elsewhere.

Does Anthropic offer HIPAA business associate agreements?

Anthropic publishes a support article on Business Associate Agreements for commercial customers, with coverage listed by product surface, so eligibility depends on which Claude product and features you use. Read that article and ask Anthropic sales before putting regulated health data through a given surface. For security certifications, use the company trust center as the authority, because certification scope and dates change.

About Anthropic

Anthropic is an AI safety company founded in January 2021 by Dario Amodei, Daniela Amodei and five other former OpenAI researchers. It sells Claude, a family of large language models, through the Claude API, Claude.ai subscriptions, Claude in Amazon Bedrock and Google Cloud Vertex AI. This record covers the Anthropic platform as a service; the corporate profile sits on the Anthropic company page.

Read the full background

The current lineup on Anthropic's pricing page has four models: Fable for demanding reasoning, Opus for long-running agentic coding, Sonnet as the balanced default and Haiku for classification, extraction and routing. Haiku is priced by prompt length, so very long prompts cost more per token. The other current models include the full one-million-token context window at standard rates, so a very long request is billed at the same per-token rate as a short one. Older generations stay listed at their earlier prices, which is useful when a workload is pinned to a model that has not been migrated.

The Batch API takes half off both input and output tokens for asynchronous work. Prompt caching charges a premium to write and a small fraction of the input price to read, with cheaper reads on the newest models, and the discount stacks with batch pricing. Web search is billed per thousand searches on top of tokens. Choosing US-only inference adds a surcharge on recent models. Billing through the API is pay as you go, with no minimum commitment on direct billing.

For people rather than programs, Claude.ai has a free plan, Pro, Max, a Team plan for small and mid-sized organizations and Enterprise, which adds usage at API rates to a per-seat fee. Claude Code is included from Pro upward; compare the agent on its Claude Code tool page and the consumer app on the Claude tool page. Exact plan prices sit in the pricing notes below, read from Anthropic's own pages.

Who should look elsewhere: teams that must run models on their own hardware cannot do so, because all inference runs on Anthropic or cloud-partner infrastructure, and generation-heavy workloads feel the gap between input and output rates. A buyer who mainly wants the lowest per-token price will find cheaper options in the directory of inference services, and Anthropic's Haiku tier is the closest it comes to competing on cost.

Sources and guides

More AI Infrastructure Services on Hokai