Claude Code
Anthropic's official terminal agent for Claude. Ships as a CLI installable from npm, with first-class support for MCP servers, custom hooks, slash commands, subagent orchestration, and a project-level CLAUDE.md memory file. The harness benefits from being co-designed with Claude's agentic post-training, which is a large part of why its published scores with Opus 4.7 and Sonnet 4.6 sit near the top of agentic-coding leaderboards. Claude Opus 5.5, its default model since September 2026, also leads Vals' Terminal-Bench 4.0 board (65.15 on the October 18, 2026 update), though Vals runs every model in the neutral mini-swe-agent harness rather than in Claude Code.
Leaderboard Placements
Live board as of 2026-10-28: federated Terminal-Bench rows, with TensorFeed editorial scores for harnesses the board does not cover.
| Benchmark | Best base model | Score | Rank |
|---|---|---|---|
| SWE-bench Verified | - | - | - |
| Terminal-Bench | - | - | - |
| Terminal-Bench 2.1 | Claude Fable 5 | 83.8 | #2 / 12 |
| Terminal-Bench 3.0 | Claude Fable 5 | 34.0 | #2 / 11 |
| Terminal-Bench 4.0 | Claude Opus 5.5 | 64.8 | #1 / 21 |
| Aider Polyglot | - | - | - |
| SWE-Lancer | - | - | - |
Distribution
CLI installable via npm. Also available as a VS Code and JetBrains extension, plus a desktop app.
Model Story
Anthropic models only. Routes Claude Opus 5.5 by default, with other Claude models selectable. No bring-your-own-key for non-Anthropic models.
Pricing
Subscription tied to Anthropic API credits or a flat-rate Pro/Max plan. No per-tool fees.
Who It's For
Engineers running long, autonomous coding sessions in a terminal where MCP integrations and project-aware memory pay off.
Notable Features
- Native MCP server orchestration (single config, hot reload)
- CLAUDE.md per-project and ~/.claude/CLAUDE.md global memory
- Hooks: pre-tool, post-tool, on-stop
- Subagent delegation with isolated worktrees
- Plan mode with approval gating