Vibeleaderboard
Anthropic

ai lab

Also indexed as claude · anthropic-ai

Anthropic

Anthropic matters because it is simultaneously a frontier model company, a leading supplier of coding agents, and one of the field's most visible experiments in turning safety research into company policy. Its decisions reveal whether those three roles reinforce one another or eventually collide.1,6,8

Official site ↗Models in the Atlas →Claude on X ↗Anthropic on X ↗Anthropic on GitHub ↗
Overall Viber #1

26 tools · 57 Intel

Profile

Overview

A safety team becomes a company

Anthropic is a San Francisco AI company founded in 2021 by seven former OpenAI employees, including siblings Dario and Daniela Amodei. The founding group had worked across language-model research, safety, policy, and operations. They organized Anthropic as a public benefit corporation and built a Long-Term Benefit Trust intended to give independent trustees a role in governance as the systems become more capable.1,9

Claude becomes a work platform

Claude is both the company's model family and its route to market. Anthropic sells access through Claude.ai, developer APIs, enterprise plans, and cloud partnerships with Amazon Web Services and Google Cloud. Claude Code turned the models into a widely used coding agent, while connectors, plugins, and industry packages widened the product from chat into software development and knowledge work.1,7

The research program

Safety research is not a side program at Anthropic. Constitutional AI trains a model to critique and revise its behavior against written principles. Mechanistic interpretability work tries to identify internal features and circuits rather than judging a model only by its outputs. System cards publish capability and risk evaluations for major releases, while the Responsible Scaling Policy links stronger capabilities to additional deployment and security controls.2,6,5

The constraint is being tested

The difficult question is whether those practices remain binding as Anthropic competes at frontier scale. In 2026 the company revised its Responsible Scaling Policy and removed an earlier commitment to stop training when safeguards lagged, replacing it with risk reporting and roadmaps. Anthropic has also withheld a stronger internal cyber-capable system while continuing a fast commercial release cadence. The company is therefore a live test of whether safety institutions can constrain a successful frontier business from inside it.8,10

Company evidence

Epoch AI dataset ↗
Reported revenue
$65BAnnualized run rate
Jul 31, 2026 · Likely[1]
Latest funding
$65B$965B post-money valuation
May 28, 2026 · Confident[1]
Reported staff
3KFull company
Aug 6, 2026 · Likely[1]
Reported compute
$4.1BR&D cloud compute · Year
Dec 31, 2025 · Likely[1]

Reported estimates, not audited figures. Confidence labels and source links are preserved from the dataset.

Reporting and context

Notable contributions

  1. 01Constitutional AIAnthropic formalized a method for supervising models with written principles and model-generated critiques, reducing dependence on direct human labels for every behavior.2
  2. 02Responsible scaling commitmentsIts Responsible Scaling Policy made capability thresholds, deployment controls, and security requirements part of a public operating framework for a frontier lab.4,5
  3. 03Detailed frontier system cardsAnthropic made model-specific capability, alignment, cyber, and misuse evaluations a visible companion to major Claude releases.6
  4. 04Interpretability at production-model scaleThe company invested in mapping learned features inside large deployed models, connecting mechanistic interpretability research to concrete model behavior.3

Company work

Built and published by Anthropic

Directly attributed work from the VibeLeaderboard catalog. This is separate from reporting and commentary about the lab.

Tools

26
Show 22 more

Intel

23
Show 19 more