Anthropic publishes selected growth and product figures, but not a complete, continuously updated set of financial statements. Realized revenue, annualized run rate, projections, and external usage data are different metrics.
- A confidential IPO prospectus reviewed by Reuters in September 2026 reported about $4.6 billion in Anthropic revenue for 2025 and an operating loss above $8 billion.
- Anthropic's latest official company-wide run-rate figure was more than $47 billion in early May 2026. Bloomberg reported more than $65 billion at the end of July based on anonymous sources and cited a New York Times projection in September of an annualized 2026 revenue pace above $100 billion.
- Anthropic's officially announced post-money valuation after its May 2026 Series H round was $965 billion. A potential IPO target above $2 trillion comes from a Reuters report based on a confidential prospectus it reviewed.
- Similarweb’s September 2026 analysis puts claude.ai at roughly 950 to 969 million visits per month, without assigning individual months to that range. The historical April figure is 824 million. These are website sessions, not active users.
- Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5, and Mythos 5.1 each have 1 million tokens of context and 128K maximum output. Opus, Sonnet, and Haiku 5.5 reach up to 300K on the Batch API with a beta header.
1. Anthropic and Key Milestones
Anthropic was founded in 2021 by former OpenAI researchers. Dario Amodei and Daniela Amodei are among its co-founders. Claude was publicly introduced in March 2023. Since then, Anthropic has added the Haiku, Sonnet, Opus, and Fable model lines; Claude Code became generally available in May 2025.
This selection brings together product and company milestones. The figures come from announcements made at different times and should not be read as a continuous series.
Anthropic introduced Claude publicly on March 14, 2023. The company describes itself as a Public Benefit Corporation. The first Claude announcement, the Series A announcement, the Claude 4 announcement that made Claude Code generally available on May 22, 2025, the announcements for the Claude Partner Network on March 12 and Claude Frontier Academy on October 2, and the model announcements for Fable and Mythos 5.1, Opus 5.5, and Sonnet 5.5 support these milestones. Anthropic announced a $100 million commitment for each of the Partner Network and Frontier Academy; the Academy announcement says it builds on the Partner Network. The sources do not clarify whether the budgets overlap, so the amounts are not added here. Both are program investments, not funding rounds.
2. Funding and Valuation
The latest publicly announced round is Series H, dated May 28, 2026. Anthropic reported a $65 billion round at a $965 billion post-money valuation. The announcement says the round includes $15 billion in previously committed hyperscaler investments, including $5 billion from Amazon.
Selected funding rounds
Terms announced by Anthropic for Series E through H.
These are announced terms for selected rounds, not a complete cash-flow statement or the sum of all financing commitments. In March 2025, Anthropic announced Series E at $3.5 billion and $61.5 billion post-money. For Series F, it announced $13 billion at $183 billion post-money. Its Series G was $30 billion at $380 billion post-money. For Series H, it announced $65 billion and $965 billion.
IPO filing and reported valuation target
On June 1, 2026, Anthropic said it had confidentially submitted a draft Form S-1 registration statement to the U.S. Securities and Exchange Commission. That gives the company the option to go public after SEC review, but does not guarantee an IPO. Anthropic said at the time that neither the number of shares nor the offering price had been set.
On September 28, 2026, Reuters reported from a confidential prospectus it had reviewed that Anthropic could target a valuation above $2 trillion. This is a media report about a potential IPO, not a completed funding valuation or confirmed market capitalization. Anthropic’s S-1 announcement and Reuters’ September 28 report document these different levels of evidence.
3. Revenue: Realized Results, Run Rate, and Reports
Revenue actually earned during a financial year is different from an annualized run rate. A run rate projects a shorter, recent period over a year. It is not automatically calendar-year revenue or the same thing as ARR.
Latest reported full-year revenue
On September 28, 2026, Reuters reported from Anthropic’s confidential IPO prospectus that the company recorded about $4.6 billion in revenue for 2025. Its operating loss exceeded $8 billion. Reuters said the prospectus’s roughly $42 billion net loss included a non-cash valuation adjustment of about $34 billion connected with financing instruments that could later convert into shares. These figures come from a confidential document, not a publicly accessible audited annual report.
The Reuters report on the prospectus also gives preliminary information about costs and obligations. Those measures should not be confused with revenue, funding, or company valuation.
Company-wide run-rate figures
Anthropic-wide revenue figures, with their measurement periods and source type.
Anthropic published several of the run-rate figures itself; the March and July values in the table come from Bloomberg reports based on anonymous sources. On March 3, Bloomberg said the run rate had passed $19 billion, citing people familiar with the matter, and reported about $14 billion a few weeks earlier. Anthropic’s April 6 announcement said the run rate exceeded $30 billion; it did not report $30 billion of revenue for the financial year. In August, Bloomberg reported that the annualized rate had passed $65 billion by the end of July. Axios also relayed preliminary quarterly figures from documents reviewed by Bloomberg, including more than $11.5 billion in second-quarter 2026 revenue. That was a preliminary quarterly figure, not a full-year financial statement.
On September 18, Bloomberg cited a New York Times report projecting that Anthropic’s annualized revenue pace would exceed $100 billion in 2026. This projection is not an audited or Anthropic-published 2026 figure. Sources for the series include the Series F announcement, Series G, Bloomberg’s March report on the run rate above $19 billion, Anthropic’s April 6 announcement, Series H, Axios on Bloomberg’s July figure, and Bloomberg’s September 18 report citing the New York Times projection.
4. Usage and Comparison Data
Anthropic does not publish a global count of active Claude accounts. The following measurements each cover their own sample or data source. Website visits, conversations, paying businesses, and API usage are different units.
Visits to claude.ai
Similarweb estimated that monthly visits to claude.ai rose from 203 million in February 2026 to 824 million in April 2026. Its later September analysis reports roughly 950 to 969 million visits per month. The public text does not assign individual months to that later range. A visit is a session; one person can generate several visits.
Similarweb’s September 2026 analysis also compares visits to major AI chatbot websites. Claude’s share rose from about 2% to 9.6% in the 12 months through August 2026, while ChatGPT’s fell from 78% to 57%. The denominator covers six websites, claude.ai, chatgpt.com, gemini.google.com, deepseek.com, grok.com, and perplexity.ai. These are shares of website visits, not users, AI spending, or revenue.
A broader Similarweb traffic breakdown for August 2026 also includes Copilot and other websites. It puts Claude at 9.3% and ChatGPT at 55.5%. Shares with different comparison groups cannot be treated as the same measurement.
Paid business use
The Ramp AI Index is based on transactions on Ramp’s platform. In August 2026, 43.8% of U.S. businesses in its data paid for Anthropic subscriptions or tokens; 39.8% paid for OpenAI. Ramp published these figures on September 9, 2026. They describe businesses in Ramp data, not a representative survey of all U.S. companies.
Share of U.S. businesses represented in Ramp data that paid each provider in August 2026.
Ramp AI Index, published September 9, 2026.
Geographic distribution: an older conversation sample
The latest complete worldwide country breakdown published in an Anthropic source is from the September 2025 Economic Index. Anthropic analyzed 1 million Claude.ai Free and Pro conversations from August 4 to 11, 2025, covering more than 150 countries. In this sample, 21.6% of conversations were from the United States, 7.2% from India, and 3.7% from Brazil.
The unit was a conversation, not a person. These are historical shares from a sample analyzed by Anthropic and do not show today’s active-user count. Anthropic published the sample in its September 2025 Economic Index.
5. The Claude Model Family
Anthropic’s model names are organized into the Fable, Opus, Sonnet, and Haiku lines. Mythos is a model with restricted access. Anthropic does not reliably publish parameter counts for these models.
Current Claude models and API limits, checked October 7, 2026. Status, release date, context window, and reliable knowledge cutoff use the project's centralized model data; output limits use centrally recorded provider facts.
Anthropic positions Fable 5.1 for demanding reasoning and long-horizon agent work. Opus 5.5 is designed for agentic coding and knowledge work. Sonnet 5.5 aims to combine speed and capability for well-scoped everyday tasks. Haiku 5.5 targets high-volume, latency-sensitive tasks such as classification, extraction, and routing. Anthropic says Mythos 5.1 shares the model with Fable 5.1 but uses different safeguards and is available only through trusted access programs.
Haiku 5.5 was released on October 7, 2026 and costs from $0.10 input and $0.50 output per million tokens according to Anthropic, with 1 million tokens of context and 128K output. The current model table and the announcements for Fable 5.1 and Mythos 5.1, Opus 5.5, and Sonnet 5.5 document model names, availability, and context windows. The table draws current context-window and output-limit values from centralized project data. Anthropic lists a regular maximum output of 128K tokens for Fable 5.1, Opus 5.5, and Sonnet 5.5, and also 128K for Haiku 5.5. Its Mythos 5.1 model page says it has the same specifications as Fable, including 128K output. The Opus documentation and Sonnet documentation separately list up to 300K output tokens on the Batch API with a beta header. That is a beta limit, not the regular output limit; context window is a separate bound.
6. Benchmarks: Provider-Reported Results
Anthropic publishes benchmark figures on each model page. This selection focuses on Terminal-Bench 4.0, an evaluation of multi-step work in a command-line environment.
Terminal-Bench 4.0 results published by Anthropic. Test conditions differ.
These are Anthropic-reported scores, not an independent overall ranking or a measure of Claude-user productivity. Effort settings and safeguards differ. Anthropic also says Fable 5.1 and Mythos 5.1 are the same underlying model with different safeguards. The Fable 5.1 and Mythos 5.1 announcement, along with the Opus 5.5 and Sonnet 5.5 pages, explains the measurement conditions.
7. Claude Code as a Separate Product
Claude Code became generally available in May 2025. On February 12, 2026, Anthropic said Claude Code alone had reached a run rate above $2.5 billion. The company said the product run rate had more than doubled since January 1, and weekly active Claude Code users had also doubled.
Anthropic also said business subscriptions had quadrupled since the start of the year and enterprise use had grown to more than half of Claude Code revenue. These are company figures dated February 2026. They should not be added to or confused with Anthropic-wide run rate. These figures appear in Anthropic’s February 12, 2026 Series G announcement.
8. API Pricing and Subscriptions
API base prices are in U.S. dollars per 1 million tokens. Input and output are billed separately. The table uses the project’s centralized price data, checked against Anthropic’s current pricing documentation.
Anthropic API base prices per 1 million tokens, in USD, as of October 7, 2026.
Prompt caching has separate write and read prices. The charge depends in part on how long the cached content is stored.
Prompt-caching prices per 1 million tokens, in USD, as of October 7, 2026.
Other terms also affect the bill. The Batch API cuts input and output prices by 50%; Anthropic says this discount stacks with prompt-caching and inference-geography modifiers. Fast mode is a research preview for Opus 5.5 and costs $8 input and $40 output per million tokens on the first-party Claude API. It is unavailable on Claude Platform on AWS, partner-operated cloud platforms, and the Batch API. Server-side web search adds $10 per 1,000 searches plus the token costs of search content. Anthropic says the full 1-million-token context window is included at standard prices for Claude 4.6 and later. For U.S.-only inference on Claude 4.6 and later, Anthropic lists a 1.1× multiplier across all token categories, including cache writes and reads; partner-cloud tariffs may differ.
Anthropic’s consumer pricing page lists these U.S. prices. Applicable taxes are excluded; usage limits apply, and prices may change.
U.S. list prices on Anthropic's pricing page, checked October 7, 2026.
Anthropic’s pricing page and its API pricing documentation are the primary references. The displayed Pro equivalent of about $17 per month is a rounded presentation of the $200 annual prepayment, not a separate monthly rate.
9. Enterprise Customers, Compute, and Published Safety Policies
Customers and infrastructure
Anthropic reported more than 300,000 business customers in October 2025. On February 12, 2026, the company said eight of the Fortune 10 were Claude customers and that more than 500 business customers each spent over $1 million on an annualized basis. On April 6, Anthropic reported more than 1,000 such customers. These are company-reported business measures, not counts of individually active Claude users. The Fortune 10 and more-than-500 figures are in the February 12 Series G announcement; the more-than-1,000 figure is in the April 6 compute announcement.
Anthropic has announced several compute agreements. In October 2025, it announced additional Google Cloud TPU capacity worth “tens of billions” of dollars. On May 28, 2026, Anthropic reported agreements for up to 5 GW of new Amazon capacity, 5 GW of TPU capacity from Google and Broadcom, and access to SpaceX GPU capacity. In April, Anthropic had also committed to spending more than $100 billion on AWS technologies over the next ten years. Capacity and spending commitments are not funding rounds or revenue. The October 23, 2025 Google Cloud announcement, the April 20, 2026 Amazon agreement, and the May 28, 2026 Series H announcement describe the agreements.
Constitutional AI and the Responsible Scaling Policy
Constitutional AI describes Anthropic’s approach to shaping model behavior with a written constitution and AI feedback. It is a training method, not a safety guarantee. Version 3.4 of the Responsible Scaling Policy took effect on July 8, 2026. Anthropic’s transparency page was updated on October 2 and links to the August 2026 Risk Report and model reports for the latest Claude releases.
These are company publications about its own processes and evaluations, not independent certifications. Claude’s Constitution, the Responsible Scaling Policy, and the Anthropic Transparency Hub document these methods.
10. What These Figures Cannot Tell Us
Anthropic does not publish a global active-user count, verified full-year revenue for 2026, or a representative 2026 distribution of Claude users. Anthropic does not publish a current employee count that would support a reliable revenue-per-employee calculation. Nor is there one consistent metric for comparing Claude and ChatGPT as overall “market leaders” across users, businesses, websites, API calls, and revenue.
The available figures cover different slices: revenue reported in a confidential prospectus, company run rate, model benchmarks, estimated website visits, transactions in Ramp data, and an older Anthropic conversation sample. These are dated indicators with different levels of evidence, not interchangeable market shares.
