COMPANY PROFILE / AI & MACHINE LEARNING
Anthropic
A safety-focused public benefit corporation that tiers who can use each model while securing compute across several clouds and chips
Anthropic is a model developer, and a public benefit corporation, best known for the Claude family and Claude Code. In 2026 its models turned over repeatedly: the top-tier Claude Fable 5 (June), the limited-access Mythos 5 for cyber and biology, then Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 from September. At the same time, events beyond technology have moved the business: a US government export control directive that suspended access in June, a dispute with the Department of War, and very large compute agreements with AWS, Google, Microsoft, SpaceX and others. This profile sets all of this out only as far as Anthropic's own announcements and court records confirm it. The company is private, so no audited financial statements are public.
What this article covers
- Anthropic in 30 seconds, and where it stands in AI and machine learning
- The logo and how the name is written
- Where the company sits in the AI field
- Scale
- Resources committed to AI and machine learning
- What it has done so far
- What comes next
- What the company is aiming for in AI
- Map of AI partnerships
- Technologies, products and services
- The risks this company carries
- Glossary, references and claim-to-source audit
1. Anthropic in 30 seconds, and where it stands in AI and machine learning
LEGAL NAMEAnthropic PBCA public benefit corporation
GOVERNANCEBoard plus the
Long-Term Benefit TrustDirectors are elected by stockholders and the Trust (company page)
Long-Term Benefit TrustDirectors are elected by stockholders and the Trust (company page)
LISTINGPrivateAnnounced a confidential draft S-1 on 1 June 2026
CATEGORY IN THIS FIELDModel developerSecondary: platform (API, Claude Code, the clouds)
CORE MODELSClaude 5.x family
Fable, Opus, Sonnet, HaikuMythos 5.1 is limited-access. Updated September to October 2026
Fable, Opus, Sonnet, HaikuMythos 5.1 is limited-access. Updated September to October 2026
FUNDING (COMPANY STATEMENT)Series H
US$65 billion28 May 2026; post-money valuation US$965 billion
US$65 billion28 May 2026; post-money valuation US$965 billion
REVENUE (COMPANY STATEMENT)Run-rate over
US$47 billionA May 2026 statement; not recognised revenue
US$47 billionA May 2026 statement; not recognised revenue
COMPUTETrainium, TPUs, NVIDIA GPUsClaude is offered on all three major clouds
The one thing to grasp first
What stands out about Anthropic is that the company itself grades how widely each more capable model is released. From one underlying model it makes two forms, a generally available Fable and a Mythos with safeguards relaxed for cyber and biology work, the latter limited to vetted organisations Sourced. In 2026 there was also an event with no parallel at other developers: a US government directive suspended access to two of its models (12 June to 1 July) Sourced. This article records that history as far as the company's account and court records go, without adding judgement. Because the company is private, amounts come only from its own statements, not audited numbers.
2. The logo and how the name is written
A note on names: the company is Anthropic (its legal name is Anthropic PBC, where PBC means public benefit corporation). Its models are called Claude, and Claude has lines named Fable, Opus, Sonnet, Haiku and Mythos, according to the company website. It is unrelated to OpenAI. The logo shown is the wordmark used in the header of the company's official website.
Official site: https://www.anthropic.com/3. Where the company sits in the AI field
4. Scale
This section covers the size of the company as a whole. Every figure is something Anthropic itself stated officially; no third-party estimates are used.
| Item | What it is | Notes and category |
|---|---|---|
| Organisation | A public benefit corporation. Six directors; three trustees of the Long-Term Benefit Trust | Directors: Dario Amodei, Daniela Amodei, Yasmin Razavi, Reed Hastings, Chris Liddell and Vas Narasimhan. Trustees: Neil Buddy Shah, Richard Fontaine and Ben Bernanke (company page, checked 15 October 2026) Sourced |
| Valuation | US$965 billion post-money | Company announcement of the Series H on 28 May 2026. The earlier Series G (12 February 2026) raised US$30 billion at US$380 billion Sourced |
| Funding | US$65 billion in the Series H | Led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital. Includes US$15 billion of previously committed hyperscaler investment (US$5 billion from Amazon) Sourced |
| Revenue | Run-rate over US$47 billion (May 2026) | A company statement. A run-rate annualises recent revenue and is not recognised revenue. Audited statements are not public Sourced |
| Large customers | More than 1,000 business customers each spending over US$1 million a year | Company statement of 6 April 2026 (over 500 at the time of the February Series G) Sourced |
| Employees | Could not be confirmed | Not stated in the official material checked Not yet confirmed |
A caution
The revenue and valuation figures are numbers written in company blog posts. Because the company is private, they cannot be cross-checked against SEC filings. On 1 June 2026 it announced that it had confidentially submitted a draft S-1, saying the number of shares and the price are not set; a confidential draft is not made public. No public S-1 turned up within the scope of this check Not yet confirmed.
5. Resources committed to AI and machine learning
The table below shows, for people, money, outside stakes and ties to universities or public bodies, whether Anthropic's official material confirms them.
| Type of resource | What it is | Notes and category |
|---|---|---|
| People | Headcount of researchers and staff, research sites | Headcount is not stated in the official material checked. A Singapore office (its fifth in Asia-Pacific) opened on 14 October 2026 Not confirmed |
| Money: R&D and capex | Accounting R&D expense and capital expenditure | Not disclosed. Contract and commitment values are separate (next row) Not disclosed |
| Money: compute contracts | Amazon: more than US$100 billion over ten years, up to 5 GW (April 2026). Microsoft and NVIDIA: US$30 billion of Azure capacity. Fluidstack: a US$50 billion investment in US AI infrastructure (announced November 2025) | Company announcements. These are contract and commitment values, not amounts already spent Sourced (contract value) |
| Outside investment | The Series H (US$65 billion) and Series G (US$30 billion). Amazon invested US$5 billion, with up to US$20 billion more | Investment into Anthropic. Amazon's further amount is a future allowance Sourced |
| Acquisitions | Bun (3 December 2025), Vercept (25 February 2026) and Stainless (18 May 2026) | Announcements on the company blog. Amounts could not be confirmed Sourced |
| Ties with memory and semiconductor companies | Micron, Samsung and SK hynix are named as strategic infrastructure partners in the Series H post | Whether they invested, and how much, is not stated Sourced |
| Universities, research bodies and public funding | US$150 million over three years to the Genesis Mission (Claude for more than 15 agencies). 10,000 free or discounted seats for academic scientists. An MOU with the Singapore government | Company programmes. No government funding was confirmed Sourced |
Status of principal products, models and facilities
Prices are the standard rates on the official pages as checked up to 15 October 2026. Prices change often, so check the current official page before relying on them.
| Product, model or facility | Scale or specification (company's figures) | Stage | Date | Source |
|---|---|---|---|---|
| Claude Fable 5.1 | API price: US$10 input / US$50 output per million tokens; cache reads US$0.25 | Released (generally available) | 1 September 2026 | Reference 10 |
| Claude Mythos 5.1 | The same model as Fable 5.1 with different safeguards. Limited to vetted organisations | Limited availability | 1 September 2026 | Reference 10 |
| Claude Opus 5.5 | US$4 input / US$20 output; cache reads US$0.20. The company says it costs 40% less than Opus 5 on typical workloads | Released | 22 September 2026 | Reference 11 |
| Claude Sonnet 5.5 | US$2 input / US$10 output. Cache reads cut to US$0.10 on 7 October | Released | 28 September 2026 | References 12 and 13 |
| Claude Haiku 5.5 | US$0.10 input / US$0.50 output (prompts up to 100,000 tokens) | Released | 7 October 2026 | Reference 13 |
| Use of Trainium | Over one million Trainium2 chips in use. About 1 GW of Trainium2 and Trainium3 planned to be online by the end of 2026 | Contracted; coming online in stages | 20 April 2026 | Reference 5 |
| Google TPUs | Multiple gigawatts of next-generation TPU capacity with Google and Broadcom | Contract (expected online from 2027) | 6 April 2026 | Reference 4 |
| SpaceX Colossus 1 | All of its capacity: over 300 MW and over 220,000 NVIDIA GPUs | Contract (said to come online within a month) | 6 May 2026 | Reference 6 |
6. What it has done so far
| When | What happened | Why it matters |
|---|---|---|
| December 2025 | Donated MCP to the Agentic AI Foundation under the Linux Foundation. Acquired Bun. Said Claude Code reached US$1 billion in run-rate revenue in six months | Moved an agent-interoperability standard to an industry body |
| 12 February 2026 | Series G: US$30 billion raised (valuation US$380 billion) | The previous round |
| 6 and 20 April 2026 | Announced multi-gigawatt TPU capacity with Google and Broadcom, and up to 5 GW (more than US$100 billion over ten years) with Amazon | Compute spread across three tracks |
| 6 May 2026 | Agreement to use all of SpaceX's Colossus 1 capacity; raised usage limits | NVIDIA GPUs also in use |
| 28 May 2026 | Series H: US$65 billion at a post-money valuation of US$965 billion | The latest financing |
| 1 June 2026 | Announced the confidential submission of a draft S-1 | Keeps the option of listing |
| 9 June to 1 July 2026 | Announced Fable 5 and Mythos 5 (9 June). A US government export control directive suspended access (12 June). The directive was lifted (30 June) and access restored on 1 July | The company's account. See also Chapters 7 and 11 |
| 20 July 2026 | The court gave final approval to the settlement of the Bartz copyright class action (US$1.5 billion fund) | Court record |
| 1 September to 7 October 2026 | Fable 5.1 and Mythos 5.1 (1 September), Opus 5.5 (22nd), Sonnet 5.5 (28th), Haiku 5.5 (7 October) | Every tier updated in about five weeks |
| 25 September 2026 | The D.C. Circuit denied the petitions challenging the Department of War's exclusion of Claude from its supply chain (the panel was divided) | Court record. A petition for rehearing was filed on 13 October |
| 2 to 14 October 2026 | Claude Frontier Academy (US$100 million), the Cyber Mission, US$150 million for the Genesis Mission, the Singapore office | Programmes for enterprises and public bodies |
7. What comes next
GOOGLE TPUSOnline from 2027 (expected)Multiple gigawatts with Google and Broadcom
TRAINIUMAbout 1 GW by the end of 2026 (planned)Trainium2 and 3; within an allowance of up to 5 GW
ENTERPRISE FRONTIER SAFEGUARDSPhased from "later this fall"Keeps data in cloud infrastructure the customer controls
EMBEDDED EVALUATORSTo be invited "in the near future" (a commitment)Stated in the CEO's essay; timing not set
All four items above are company plans or commitments, not things that have happened Not yet confirmed. The same applies to the items below.
| Plan | What the company states | Category |
|---|---|---|
| Claude Frontier Academy | To train 10,000 "Frontier Deployed Engineers" by the end of 2027 (US$100 million) | Not yet confirmed |
| Expanding the Cyber Verification Program | Advanced cyber capabilities for verified defenders on Opus 5.5, Sonnet 5.5, Mythos 5.1 and others (Project Glasswing merged in) | Not yet confirmed |
| Amazon's further investment | Up to US$20 billion (future) | Not yet confirmed |
| Listing | The S-1 was submitted confidentially. Timing and price not set; the company says it depends on market conditions and other factors | Not yet confirmed |
| Pacing the frontier, as asked of others | The CEO proposes three steps: embedded third-party evaluators, coordination among democracies, and global coordination. Whether this happens is not settled | Not yet confirmed |
8. What the company is aiming for in AI
This section goes beyond what the company says directly. It sets out inferences that official material supports, each with its basis.
| What can be read | Primary evidence behind it | Category |
|---|---|---|
| It is trying to reconcile capability and safety by tiering release | The same model is split into Fable (general) and Mythos (limited), with Mythos restricted to trusted organisations. The Fable 5.1 post also addresses reducing safeguard false positives | Sourced |
| It spreads across the three major clouds and three kinds of chip to make supply harder to interrupt | Offered on AWS, Google Cloud and Azure; the company says it uses Trainium, TPUs and NVIDIA GPUs, and calls AWS its primary cloud and training partner | Sourced |
| It is widening enterprise and developer use while cutting prices | Opus 5.5 is priced 20% lower per token than Opus 5 and described as 40% cheaper on typical workloads. Sonnet 5.5's cache price was cut in October | Inference |
| It treats slowing the pace of development itself as a safety measure | The CEO's essay proposes "pacing" and commits the company alone to embedded evaluators. New models continued to ship over the same period | Inference |
| In contracts with the public sector it kept bans on use for lethal autonomous warfare and for mass surveillance of Americans | The D.C. Circuit opinion says the department decided on the exclusion after the company did not relax those two prohibitions | Sourced |
One inference, explained in full
In the Opus 5.5 announcement the company says that at these capability levels benchmark margins have become a less reliable guide to real-world differences, and shifts its emphasis to cost per task and completion of real work Sourced. The numbers on the same page are framed as "performance at the level of Fable 5.1 at 40% lower cost". From this, this article reads the company as moving its pitch from a contest on peak capability toward cost per unit of performance and an account of safety Inference. That reading is an interpretation of public posts and does not assert the company's intent.
9. Map of AI partnerships
10. Technologies, products and services
These are the main products and mechanisms on offer.
| Product or mechanism | What it is (as stated in official announcements, checked 15 October 2026) | Source |
|---|---|---|
| Claude (chat and apps) | Conversation and work products for consumers and businesses. Fable 5.1 is available on Pro, Max, Team and Enterprise | Reference 10 |
| Claude Code | A coding agent. The company says it reached US$1 billion in run-rate revenue in November 2025. Claude Code Enterprise also exists | Reference 21 |
| Claude Platform (API) | For developers. Also offered on AWS (Bedrock), Google Cloud (Vertex AI) and Microsoft (Foundry) | Reference 4 |
| Claude Cowork, Claude Science, Claude Security and others | Listed in the product menu on the company website (checked 15 October 2026). Specifications and terms were not checked for this article | Reference 1 |
| Model Context Protocol (MCP) | An open standard connecting agents to outside tools. Donated to the Agentic AI Foundation in December 2025 | Reference 20 |
| Responsible Scaling Policy | The company's policy of scaling safeguards to severe risks. Version 3.4 took effect on 8 July 2026 | Reference 14 |
How to read the company's own evaluations
The Opus 5.5 page carries a table setting its own and other vendors' models (such as OpenAI's GPT-6 Astra) side by side. Anthropic ran the evaluations in its own environment and quotes other vendors' values from their published figures or public leaderboards. Settings (reasoning effort, harness) differ by evaluation, and the company itself notes that the gaps reflect real-world differences poorly, so this article does not reproduce the scores or rank anyone against anyone Sourced.
A materials engineer's view: how widely to release is decided by both safeguards and compute
With Fable 5 the company explains that when safeguards trigger, another model answers instead (initially Opus 4.8) Sourced. A design in which classifiers examine inputs and outputs and sometimes route to another model adds extra computation and latency, which feeds straight into serving cost. The cuts in cache-read prices since September 2026, and the option to keep data in the customer's own cloud, can be read as absorbing that burden in price and operations Inference.
On compute, the company says it uses Trainium, TPUs and NVIDIA GPUs and matches workloads to the chips best suited Sourced. Training and inference are limited by different things: memory bandwidth, chip-to-chip communication and power. See also our explainers on AI accelerators, training clusters and interconnects, and HBM.
11. The risks this company carries
| Risk | Substance | Category |
|---|---|---|
| Finances cannot be verified externally | Run-rate revenue, valuation and contract values come only from company announcements. Audited statements are not public and no S-1 has been published | Not yet confirmed |
| Suspension under a government directive | On 12 June 2026 a US government export control directive required access to Fable 5 and Mythos 5 to be cut off for foreign nationals, and the company suspended all customers. The controls were lifted on 30 June. The company says the letter gave no detail of the concern and that the vulnerabilities found with the technique cited were minor | Sourced |
| Exclusion by the Department of War | On 25 September 2026 the D.C. Circuit denied the petitions challenging the department's decision to exclude Claude from its supply chain (one judge dissented). The company filed a petition for rehearing en banc on 13 October. No outcome is predicted here | Sourced |
| Copyright and patent litigation | The Bartz book-authors class action settled for a US$1.5 billion fund, with final approval on 20 July 2026, though an objector filed a notice of appeal on 31 August. Concord Music Group v. Anthropic (5:24-cv-03811), brought by music publishers, and several other cases are pending. No outcome is predicted here | Sourced |
| Very large compute commitments | More than US$100 billion to Amazon over ten years, US$30 billion of Azure and more. Whether demand and utilisation match them cannot be confirmed from this material | Inference |
| Safeguard false positives | The company says Fable 5's safeguards sometimes block harmless requests, triggering in under 5% of sessions on average, and says it aims to reduce this in Fable 5.1 and Opus 5.5 | Sourced |
| Fast model turnover | Every tier updated in about five weeks from 1 September to 7 October 2026. Prices and availability change quickly | Sourced |
12. Glossary
- Public benefit corporation (PBC)
- A company form that runs a for-profit business but is required by its charter to pursue public benefit too.
- Long-Term Benefit Trust
- A trust involved in electing some of Anthropic's directors. Its trustees are listed on the company page.
- Fable / Mythos
- Two forms of the same underlying model. Fable is generally available; Mythos is limited-access with safeguards relaxed for cyber and biology work (Anthropic's naming).
- Run-rate
- Recent revenue annualised. It is not recognised annual revenue.
- Export control directive
- A government order restricting foreign nationals' access to technology or services on national-security grounds.
- Responsible Scaling Policy
- Anthropic's policy of strengthening safeguards in steps as capability rises.
- MCP
- Model Context Protocol, an open standard that lets AI agents connect to outside tools and data.
- Trainium / TPU
- AI chips designed by AWS and by Google respectively.
- Rehearing en banc
- A rehearing by all (or a majority) of an appeals court's judges. A petition is not necessarily granted.
13. References
- Anthropic Company page. That it is a public benefit corporation, and the composition of the board and the Long-Term Benefit Trust. Checked on 15 October 2026. https://www.anthropic.com/company
- Anthropic Anthropic raises $65B in Series H funding at $965B post-money valuation (28 May 2026). Investors, US$15 billion from hyperscalers, ties with Micron, Samsung and SK hynix, and the run-rate revenue statement. https://www.anthropic.com/news/series-h
- Anthropic Anthropic raises $30 billion in Series G funding at $380 billion post-money valuation (12 February 2026). https://www.anthropic.com/news/anthropic-raises-30-billion-series-g-funding-380-billion-post-money-valuation
- Anthropic Anthropic expands partnership with Google and Broadcom for multiple gigawatts of next-generation compute (6 April 2026). TPU capacity expected from 2027, and the run-rate revenue statement of over US$30 billion. https://www.anthropic.com/news/google-broadcom-partnership-compute
- Anthropic Anthropic and Amazon expand collaboration for up to 5 gigawatts of new compute (20 April 2026). A commitment of more than US$100 billion over ten years, and Amazon's US$5 billion investment with up to US$20 billion more. https://www.anthropic.com/news/anthropic-amazon-compute
- Anthropic Higher usage limits for Claude and a compute deal with SpaceX (6 May 2026). All of Colossus 1's capacity, the US$30 billion of Azure capacity with Microsoft and NVIDIA, and the US$50 billion investment with Fluidstack. https://www.anthropic.com/news/higher-limits-spacex
- Anthropic Claude Fable 5 and Claude Mythos 5 (9 June 2026). Fable 5, a Mythos-class model made safe for general use, and the limited-access Mythos 5. https://www.anthropic.com/news/claude-fable-5-mythos-5
- Anthropic Statement on the US government directive to suspend access to Fable 5 and Mythos 5 (12 June 2026). The company's account of the export control directive and the suspension of access. https://www.anthropic.com/news/fable-mythos-access
- Anthropic Redeploying Fable 5 (30 June 2026, updated 1 July). The company's account of the lifting of the export controls and redeployment. https://www.anthropic.com/news/redeploying-fable-5
- Anthropic Introducing Claude Fable 5.1 and Claude Mythos 5.1 (September 2026). Fable 5.1 pricing, Enterprise Frontier Safeguards and the limited availability of Mythos 5.1. https://www.anthropic.com/introducing-claude-fable-mythos-5-1
- Anthropic Claude Opus 5.5 (22 September 2026). Pricing, the company's own evaluations, and its remark that benchmark margins are a less reliable guide to real-world differences. https://www.anthropic.com/claude-opus-5-5
- Anthropic Claude Sonnet 5.5 (28 September 2026). Pricing and availability across platforms. https://www.anthropic.com/claude-sonnet-5-5
- Anthropic Claude Haiku 5.5 (7 October 2026). Pricing, and the cut in the cache-read price of Sonnet 5.5. https://www.anthropic.com/claude-haiku-5-5
- Anthropic Responsible Scaling Policy (updated 14 August 2026). The version history (3.4 effective 8 July 2026) and risk reports. https://www.anthropic.com/responsible-scaling-policy
- Dario Amodei We Must Pace the Frontier (September 2026). An essay by the company's CEO, stating that the company commits unilaterally to embedding third-party evaluators. https://darioamodei.com/post/we-must-pace-the-frontier
- Anthropic Anthropic confidentially submits draft S-1 to the SEC (1 June 2026). The confidential submission and the fact that share count and price are not set. https://www.anthropic.com/news/confidential-draft-s1-sec
- Anthropic Building on our commitment to American scientific discovery (8 October 2026). A three-year US$150 million commitment to the Genesis Mission. https://www.anthropic.com/news/genesis-mission-commitment
- Anthropic Anthropic invests $100 million to train 10,000 engineers and tackle the enterprise AI talent gap (2 October 2026). Claude Frontier Academy. https://www.anthropic.com/news/claude-frontier-academy
- Anthropic Introducing the Anthropic Cyber Mission (8 October 2026). Critical infrastructure defence, OSS Scanner, and the merger of Project Glasswing into the Cyber Verification Program. https://www.anthropic.com/news/anthropic-cyber-mission
- Anthropic Donating the Model Context Protocol and establishing the Agentic AI Foundation (9 December 2025). The donation of MCP to a fund under the Linux Foundation. https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation
- Anthropic Anthropic acquires Bun as Claude Code reaches $1B milestone (3 December 2025). https://www.anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone
- Anthropic Anthropic acquires Vercept to advance Claude's computer use capabilities (25 February 2026). https://www.anthropic.com/news/acquires-vercept
- Anthropic Anthropic acquires Stainless (18 May 2026). https://www.anthropic.com/news/anthropic-acquires-stainless
- Anthropic Anthropic opens Singapore office and signs MOU with the Government of Singapore (14 October 2026). https://www.anthropic.com/news/singapore-office-opening
- U.S. District Court, Northern District of California Docket of Bartz v. Anthropic PBC, Case No. 3:24-cv-05417-AMO (via CourtListener), including Dkt. 680 (20 July 2026, final approval of the settlement and judgment). https://www.courtlistener.com/docket/69058235/bartz-v-anthropic-pbc/
- U.S. Court of Appeals for the D.C. Circuit Docket of Anthropic PBC v. United States Department of War, No. 26-1049 (consolidated with 26-1162) (via CourtListener): the opinion and judgment of 25 September 2026 and the petition for rehearing en banc of 13 October. https://www.courtlistener.com/docket/73496372/anthropic-pbc-v-united-states-department-of-war/
- U.S. District Court, Northern District of California Docket of Concord Music Group, Inc. v. Anthropic PBC, Case No. 5:24-cv-03811 (via CourtListener). https://www.courtlistener.com/docket/68889092/concord-music-group-inc-v-anthropic-pbc/
14. Claim-to-source audit
| Claim in the text | Category | Source |
|---|---|---|
| Anthropic is a public benefit corporation; the composition of its board and Long-Term Benefit Trust | Sourced | Reference 1 https://www.anthropic.com/company |
| Series H: US$65 billion, US$965 billion valuation, run-rate over US$47 billion, ties with memory makers | Sourced | Reference 2 https://www.anthropic.com/news/series-h |
| Series G: US$30 billion at US$380 billion (12 February 2026) | Sourced | Reference 3 https://www.anthropic.com/news/anthropic-raises-30-billion-series-g-funding-380-billion-post-money-valuation |
| The Google and Broadcom TPU agreement (from 2027), run-rate over US$30 billion and over 1,000 large customers | Sourced | Reference 4 https://www.anthropic.com/news/google-broadcom-partnership-compute |
| The Amazon agreement (over US$100 billion over ten years, up to 5 GW) and Amazon's further investment | Sourced | Reference 5 https://www.anthropic.com/news/anthropic-amazon-compute |
| SpaceX's Colossus 1, US$30 billion of Azure with Microsoft and NVIDIA, and US$50 billion with Fluidstack | Sourced | Reference 6 https://www.anthropic.com/news/higher-limits-spacex |
| The Fable 5 and Mythos 5 announcement and the mechanism by which another model answers when safeguards trigger | Sourced | Reference 7 https://www.anthropic.com/news/claude-fable-5-mythos-5 |
| The company's account of the export control directive of 12 June 2026 and the suspension | Sourced | Reference 8 https://www.anthropic.com/news/fable-mythos-access |
| The company's account of the lifting of controls on 30 June and redeployment on 1 July | Sourced | Reference 9 https://www.anthropic.com/news/redeploying-fable-5 |
| Fable 5.1 and Mythos 5.1 pricing, availability, and Enterprise Frontier Safeguards | Sourced | Reference 10 https://www.anthropic.com/introducing-claude-fable-mythos-5-1 |
| Opus 5.5 pricing and the note that benchmark gaps reflect real-world use poorly | Sourced | Reference 11 https://www.anthropic.com/claude-opus-5-5 |
| Sonnet 5.5 pricing and platforms | Sourced | Reference 12 https://www.anthropic.com/claude-sonnet-5-5 |
| Haiku 5.5 pricing and the cut in Sonnet 5.5's cache price | Sourced | Reference 13 https://www.anthropic.com/claude-haiku-5-5 |
| The version history of the Responsible Scaling Policy (3.4 effective 8 July 2026) | Sourced | Reference 14 https://www.anthropic.com/responsible-scaling-policy |
| The CEO's essay on the company's unilateral commitment to embedded third-party evaluators | Sourced | Reference 15 https://darioamodei.com/post/we-must-pace-the-frontier |
| The confidential S-1 submission (1 June 2026) and that share count and price are not set | Sourced | Reference 16 https://www.anthropic.com/news/confidential-draft-s1-sec |
| US$150 million over three years to the Genesis Mission and 10,000 seats for academic scientists | Sourced | Reference 17 https://www.anthropic.com/news/genesis-mission-commitment |
| Claude Frontier Academy (US$100 million, 10,000 people by the end of 2027) | Sourced | Reference 18 https://www.anthropic.com/news/claude-frontier-academy |
| The Cyber Mission and the merger of Project Glasswing into the Cyber Verification Program | Sourced | Reference 19 https://www.anthropic.com/news/anthropic-cyber-mission |
| The donation of MCP to the Agentic AI Foundation | Sourced | Reference 20 https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation |
| The acquisitions of Bun, Vercept and Stainless | Sourced | Reference 21 https://www.anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone Reference 22 https://www.anthropic.com/news/acquires-vercept Reference 23 https://www.anthropic.com/news/anthropic-acquires-stainless |
| The Singapore office and MOU | Sourced | Reference 24 https://www.anthropic.com/news/singapore-office-opening |
| Bartz: final approval of the US$1.5 billion settlement (Dkt. 680) and an objector's notice of appeal | Sourced | Reference 25 https://www.courtlistener.com/docket/69058235/bartz-v-anthropic-pbc/ |
| The D.C. Circuit's decision of 25 September 2026 (denied; divided panel) and the petition for rehearing of 13 October | Sourced | Reference 26 https://www.courtlistener.com/docket/73496372/anthropic-pbc-v-united-states-department-of-war/ |
| Concord Music Group v. Anthropic and several other cases are pending | Sourced | Reference 27 https://www.courtlistener.com/docket/68889092/concord-music-group-inc-v-anthropic-pbc/ |
| Headcount, R&D expense and acquisition amounts | Not yet confirmed | Not stated in the official material checked, so this article says "could not be confirmed". Third-party estimates are not used |
| The reading that it reconciles capability and safety by tiering release | Inference | This article's interpretation, drawn from the Fable and Mythos release forms |
| The reading that it is moving its pitch from peak capability to cost per unit of performance and safety | Inference | This article's interpretation, drawn from the Opus 5.5 announcement |
| The risk of whether demand matches the size of the compute commitments | Inference | This article's interpretation, based on its observation of the contract values |
| Classification of the company (model developer primary, platform secondary) | Our calculation | This article's own classification |
Last updated: 15 October 2026 / Troy Technical
The figures in this article come from announcements on Anthropic's own website and from U.S. federal court records. No research-firm estimates or press-only figures have been used. Because the company is private, run-rate revenue, valuation and contract values are "figures the company stated officially", not audited numbers. Passages marked "Inference" are this article's interpretation of official material, not statements made by the company. No outcome of a government directive or lawsuit is predicted.