Claude Code vs Codex: Which Is Better? (Benchmarks + Pricing)
Facts checked
What the benchmarks showed
- Lava lamp build — Codex: about 56 minutes and 67k tokens (≈$6 at API prices). Claude Code: over an hour and 676k tokens (≈$35). Claude's result looked a bit better; Beau still picked Codex on cost.
- Landing page recreation — Codex: about 26 minutes of work and 199k tokens. Claude Code: about 45 minutes and roughly 3x the tokens (≈$30–35).
- Browser horror game — Codex: 3 minutes, a working but plain game. Claude Code: 19 minutes and ~200k tokens, and the game crashed.
- 3D donut in Blender — Claude Code's render was clearly better — the strongest win for Claude on visual quality.
- A full day of testing — Claude Code (Fable 5 + Opus 4.8) would have cost $900–1,000 at API prices; Codex (GPT-5.6) about $50.
Pricing (checked October 14, 2026): Claude Code needs a paid Claude plan — Pro at $20/month, or Max at $100 or $200/month — with a 5-hour session window plus a weekly limit. Codex is included on every ChatGPT plan, including Free and Go ($8/month), with Plus at $20 and Pro from $100; ChatGPT Pro plans currently have no five-hour limit. Both companies change limits often, so check their pricing pages before you commit.
Models move fast. These tests used the July 2026 flagships; OpenAI's Codex pricing page now lists GPT-6.1 Sol. The pattern has held across our benchmarks, though: OpenAI's agent is cheaper and faster per task, Anthropic's is stronger on visual polish and large refactors.
Our recommendation for beginners: start with Codex because the free ChatGPT plan includes it, add Claude Code when you need its strengths, and for anything important run both against the same tests and keep the better result. The benchmark videos below show every test side by side.
