HomeFull speed ahead. Keep track of advancements in every frontier
Claude CodeAnthropic—Anthropic—Claude CodeAnthropic—Claude modelsAnthropic—Claude modelsAnthropic—Claude CodeAnthropic—Anthropic—Claude modelsAnthropic—Anthropic—Claude modelsAnthropic—Anthropic—Claude modelsAnthropic—Claude modelsAnthropic—Anthropic—

Anthropic

Claude models, Claude Code, AI safety

US$965B privateest. 2021anthropic.com (opens anthropic.com)Checked 25 Sep

The brief

Analyst view
  • Raised $65B at a $965B valuation in May 2026, after a $30B round in February.
  • Claude Code, launched in 2025, made Claude a leading model for AI coding agents.
  • Trains on Google, Amazon and NVIDIA chips instead of relying on one supplier.

Technical approach

As reported
Alignment

Training with a written constitution

Claude learns to critique and fix its own answers against written principles, so fewer human ratings are needed.

Interpretability

Reading the model's mind

Anthropic finds millions of human-readable concepts inside Claude and traces how they combine into an answer.

Governance

Responsible Scaling Policy

Stricter safety levels (ASL) kick in as models get more capable. ASL-3 began with Claude Opus 4 in May 2025.

Agents

Coding and tools

Models are trained to use terminals and browsers for long tasks. Its MCP standard lets agents plug into outside tools and data.

Programs

Anthropic, two programs

Updated 18d ago
Research · Proto · Pilot · Deploy · Scaling

Research & demos

5 items
  • Mar 2025
    paperOn the Biology of a Large Language Model (circuit tracing)

    Traced how a model reasons in steps and even plans ahead when writing rhyming poems.

  • Dec 2024
    paperAlignment Faking in Large Language Models

    Showed a model playing along in training to avoid having its preferences changed.

  • Nov 2024
    codeModel Context ProtocolClaude Code

    Open standard for plugging AI agents into tools and data, now used across the industry.

  • May 2024
    paperScaling Monosemanticity: Extracting Interpretable Features from Claude 3 SonnetClaude models

    First time millions of readable concepts were pulled out of a full-size production model.

  • Dec 2022
    paperConstitutional AI: Harmlessness from AI Feedback

    Replaced much human safety labelling with the model critiquing itself against written principles.

Open technical problems

Open · Active · Partial
  • Scaling interpretabilityActive research

    Single concepts can be found, but checking a whole frontier model for hidden goals is still far off.

  • Alignment fakingActive research

    Tests showed a model can play along during training while keeping other preferences. That makes training results harder to trust.

  • Compute supplyOpen

    Fast growth needs gigawatts of new capacity across three chip types, each with its own software.

Peers

Frontier models · 5

Company & funding

6 rounds · as reported
Valuation
$965Bprivate
Raised
$131Bdisclosed private rounds
Last round
Series HMay 2026
HQ
US
Founded
2021
$10B$100B$1T20232024202520262027$62B$183B$380B$965B

Post-money valuation, log scale

DateRoundAmountPost-moneyLead
May 2026Series H$65B$965BAltimeter Capital, Dragoneer, Greenoaks, Sequoia Capital
Feb 2026Series G$30B$380BGIC, Coatue
Nov 2025Strategic investment$15B—Microsoft, NVIDIA
Sep 2025Series F$13B$183BICONIQ, Fidelity, Lightspeed Venture Partners
Mar 2025Series E$3.5B$62BLightspeed Venture Partners
Sep 2023Strategic investment (up to $4B)$4B—Amazon

Partners & customers

3 investors · 1 supplier · 1 customer
  • Investor
    Amazon

    Major investor and primary training partner; Project Rainier Trainium2 cluster came online in 2025.

  • Supplier
    Google

    Investor and TPU supplier; October 2025 deal for up to one million TPUs.

  • Investor
    Microsoft

    Invested up to $5B in Nov 2025; Anthropic committed to buy $30B of Azure compute.

  • Investor

    Invested up to $10B in Nov 2025 with a joint engineering partnership.

  • Customer
    Salesforce

    Enterprise distribution partner integrating Claude.

People

Key people
  • Dario AmodeiCEO and co-founder
  • Daniela AmodeiPresident and co-founder
  • Jared KaplanChief Science Officer and co-founder
Primer