Product family
Anthropic, product by product
Anthropic's product surface mapped by what you can do with each line: five judged products from the assistant app to Claude Design, and honest not-yet-judged cards for the rest.
5 of 13 lines judged in an arena · flagship: Claude in AI Assistants
Claude
JudgedAsk, research with citations, analyze files, and build artifacts in one assistant across web, desktop, and mobile — with memory, connectors, and voice that carry across every surface.
Claude Code
JudgedShip code end-to-end from the terminal, IDE, web, or CI: explore a codebase, edit files, run commands, and open the PR — with subagents, skills, hooks, and MCP.
Claude Design
JudgedTurn a written brief into interactive prototypes, decks, and marketing collateral: apply your own design system, refine with inline comments and sliders, and hand off to Claude Code to ship. An Anthropic Labs beta.
Claude Cowork
Not yet judgedHand Claude a folder and an outcome: it works through files, browsing, and multi-step office tasks in parallel sessions you review — the Claude Code harness for non-coders.
A surface of the Claude app subscription, not separately adopted — its delegate-whole-tasks capabilities are judged inside the claude entry's agents-tasks stories; a second row would double-list one product.
Claude Agent SDK
JudgedBuild your own agents on the same harness as Claude Code: the agent loop, tools, hooks, MCP, and skills as TypeScript and Python libraries.
Anthropic Skills
JudgedPackage instructions, scripts, and resources into SKILL.md folders that Claude loads on demand — portable across the app, Claude Code, Cowork, and the API.
Claude in Chrome
Not yet judgedClaude acts in your browser: navigate, fill forms, and run web tasks from a side panel with per-site permissions and action confirmations.
No browsers arena yet (roadmap: planned), and browser-agents ranks developer automation frameworks — judging a consumer extension there would be the wrong axis (same call as Perplexity Comet).
@Claude (Slack & Teams)
Not yet judgedTag @Claude in a Slack or Teams thread to summarize, answer with workspace context, run analysis, and carry out standing instructions.
A bot that lives inside team-chat products, not a chat platform — ranking it against Slack itself would be the wrong axis; the underlying assistant is judged as claude.
Claude for Microsoft 365
Not yet judgedWork inside Office: interrogate and update Excel models cell by cell, draft Word docs with tracked changes, build decks from your templates, and triage Outlook.
claude.com/claude-for-microsoft-365 ↗
An integration surface of the Claude subscription inside Microsoft's apps — no office-suite arena, and the underlying assistant is already judged as claude.
Claude Science
Not yet judgedA workbench for scientists: query scientific databases, run reproducible analyses on managed compute, visualize proteins and genomic data, and draft the manuscript alongside.
claude.com/product/claude-science ↗
Beta. ai-research-agents ranks literature-review agents (Elicit, Consensus, FutureHouse); a compute-and-analysis workbench is a neighboring but different job. Watchlist line.
Claude Security
Not yet judgedPoint it at a codebase to find vulnerabilities, adversarially validate which findings are real, and get patch suggestions your team reviews.
claude.com/product/claude-security ↗
Enterprise public beta; security-scanners ranks GA scanners any developer can adopt today (Semgrep, Snyk, Trivy). Bring-up candidate once generally available.
Managed Agents
Not yet judgedHost agents on Anthropic-run infrastructure straight from the API: managed sandboxes, stateful sessions, and scheduled runs without operating your own harness.
Beta surface of the developer platform; proprietary model platforms aren't an arena we rank yet (roadmap: frontier-model-apis).
Claude Developer Platform
Not yet judgedThe API platform: Messages API, tool use, code execution, files, and agent infrastructure for building on Claude.
Proprietary model APIs are not an arena we rank — inference-providers ranks hosts of open models, a different job (roadmap: frontier-model-apis).
A judged line competes in its arena on the same stories as every rival — family membership never affects scoring. Lines without a fitting arena stay unscored until one exists. See the methodology.