Claude Code is a CLI-first coding agent. You give it a goal in a repo; it plans, edits, runs commands, and iterates. It is not a chat window pasted into your editor. It is closer to a junior engineer with shell access, which is why teams either love it or ban it on production repos.
This page is the supporting note. Pricing (Pro $20/mo, no free Claude Code tier): Claude Code price. Install (npm, Node 18+, Windows/macOS/Linux): Claude Code installation. Do not search for a fourth URL if you typed cluade or npx claude-code — those intents stay on the install page.
Why people use it
- Lives in the terminal next to git — diffs, tests, and commits stay in one flow
- Pairs with Anthropic frontier models. On the 11 Aug 2026 Artificial Analysis snapshot, Claude Opus 5 (Adaptive Reasoning, Max Effort) ranks #1 at 63.1. The 16 Aug 2026 ingest in
content/leaderboard_data.jsonprints Claude Opus 5 at rank 1 / 63 (integers). We reprint; we do not run the evals. Live table: /leaderboard - Fits long, supervised refactors better than one-shot chat where you paste files back and forth
When to pick something else
- You want inline IDE edits all day → Cursor or Windsurf (see /posts/cursor-vs-windsurf)
- Your process is PR-centric on GitHub → Copilot Agent Mode
- You need the current coding-agent bench leader regardless of vendor → on 11 Aug, GPT-5.6 Sol (max) sat at #5 / 60.9; on 16 Aug the file has GPT 5 6 Sol at #5 / 61. Head-to-head: /posts/claude-opus-5-vs-gpt-5-6-sol-which-wins-for-agentic-coding
Limits
Claude Code bills through Anthropic usage. It will run commands you allow. Treat repo access like any other agent: branch protections, read-only sandboxes, and human review before merge. Swap models on /leaderboard when the board moves; you do not need to rewrite your workflow.
Startup credits that might offset API spend are on /perks. Sign in or confirm the newsletter email to see apply links. Method for scores: /methodology.