All tasks / Comparisons

Claude Code vs Codex

Compared only on public benchmark results, across the 1 task where both are listed: Claude Code is a tier higher in 1 of 1 shared task.

Claude Code

Anthropic's agentic coding tool for terminal, IDEs, desktop, web, mobile and Slack that reads, edits and runs code.

Vendor: Anthropic
Kind: Agents
Runs: Cloud
Price: From $20/mo
Skill level: Intermediate: some setup or configuration
Open source: No
Last verified: 18 Oct 2026

Alternatives to Claude Code

Codex

OpenAI's coding agent for the terminal, IDE, desktop app, web and iOS; the CLI is open source.

Vendor: OpenAI
Kind: Agents
Runs: Cloud
Price: Free tier · from $8/mo
Skill level: Intermediate: some setup or configuration
Open source: Yes
Last verified: 18 Oct 2026

Task by task

Task	Claude Code	Codex	Higher tier
Coding Coding agents	S68 on Artificial Analysis Coding Agent Index (v1.5) · Cloud	A63 on Artificial Analysis Coding Agent Index (v1.5) · Cloud	Claude Code

A tier spans a range of scores, and small gaps on a leaderboard are often within its margin of error, so only a different tier counts as ahead. How we rank.

Same benchmark, both results

Benchmark	Claude Code	Codex	Snapshot
Artificial Analysis Coding Agent Index (v1.5) Composite index: equal-weight mean of pass@1 (3 attempts/task) on DeepSWE v1.1, Terminal-Bench 4.0 and SWE-Atlas-QnA, per agent + model, higher is better	68 Claude Code + Sonnet 5.5 (max)	63 Codex + GPT-6.1 Sol (xhigh)	18 Oct 2026
Terminal-Bench 2.0 Accuracy: share of terminal tasks solved by an agent harness + model pair, higher is better	58% Claude Code + Claude Opus 4.6	82.2% Codex CLI + GPT-5.5	18 Oct 2026
Terminal-Bench 2.1 Accuracy: share of terminal tasks solved by an agent harness + model pair, higher is better	83.8% Claude Code + Fable 5	87.4% Codex + GPT-6 Astra	18 Oct 2026
Terminal-Bench 3.0 Accuracy: share of terminal tasks solved by an agent harness + model pair, higher is better	34% Claude Code + Fable 5	34.6% Codex + GPT-5.6 Sol	18 Oct 2026
Terminal-Bench 4.0 Accuracy: share of terminal tasks solved by an agent harness + model pair, higher is better	64.8% Claude Code + Opus 5.5	58.2% Codex + GPT-6 Astra	18 Oct 2026