Claude Code

Claude Code — Technology. approach: Direct search via grep/glob without indexing · strength: Instruction following, code generation, tool use · self-writing: ~90% of code written by itself · category: CLI coding agent · category: Coding agent, CLI · creation_date: September 2024 · context_window: 1 million tokens · interface: terminal-first CLI agent · business model: Proprietary, bundled with Claude Pro/Max · affected_products: Claude Code, Claude Agent SDK, Claude Cowork · revenue: $500M annual (run-rate) · annualized revenue: ~$1 billion (November 2025) · role: De facto evaluation harness of the post: version 2.1.207 is cited as a measurement dependency for Terminal-Bench 2.1 and 3.0, Agents' Last Exam, CyberGym, ExploitGym, ExploitBench, PostTrainBench and SWE-Marathon — a Chinese lab uses it to measure its own model and those of its competitors. Also cited as a compatible client for GLM-5.3 · criticized role: "Behavioral funnel" / hostage layer locking down the dev · stack: TypeScript, React terminal · usage: Coding agent used cross-functionally at Anthropic (product, devrel, design write code) for end-to-end delivery · described_usage: Execution of detailed plans in Fidelity One and Two · reported uses: Prototyping by non-developers, code review, single-purpose agents, self-service analytics, assisted rebuilds; the features cited are MCP, skills, CLAUDE.md, Code Review, hooks, dynamic workflows, plan mode · corrected_version: v2.1.116 (April 20, 2026) · versions: `/goal` shipped v2.1.139 (May 11, 2026), branch 2.1.x, Opus 4.8 by default · publisher: Anthropic