# Codex vs Claude Code: GPT-6.1 Sol vs Claude Opus 5.5 on 21 coding jobs

> A head-to-head test by The Call Center Doctors, run October 3, 2026. One run per job. Every number comes from one results file, which is downloadable.

## Setup
- Claude Code with claude-opus-5-5, effort high. Codex with gpt-6.1-sol, reasoning effort high.
- 21 small coding jobs in Node.js and Python. Same prompt, same starting code, same machine, 60-minute limit, no plug-ins, no house rules.
- Hidden tests decide pass or fail. A blind judge (Google Gemini 3.1, preview) rated each answer and picked one per pair with names scrubbed.
- One re-check (Sol's t07 answer after its review round) was judged by DeepSeek, because our first judge hit its usage limit. DeepSeek re-judged both original t07 answers too and agreed with Gemini on both.

## Key numbers
- Faster on: Opus, 21 of 21 jobs.
- Median time per job: Opus 74.3 s, Sol 315.1 s (about 4.2x).
- All 21 jobs added up: Opus 1821.1 s (30 min 21 s), Sol 8095.9 s (2 h 14 min 56 s), about 4.4x.
- Jobs with every hidden test passed, first try: Opus 20 of 21, Sol 18 of 21.
- Hidden tests passed: Opus 509 of 514, Sol 507 of 514.
- Judge approved on the first try: Opus 21, Sol 19 (one of Sol's two misses was t09, where we think the judge was wrong). After review rounds: both 21.
- Blind head-to-head picks: Opus 8, Sol 8, ties 5.
- False "done" messages: Opus 0, Sol 1 (t07, fixed in one review round). The judge also failed Sol on t09; we think the judge was wrong there (Sol matched the Python original and passed 300 of 300).
- Output tokens, all jobs: Opus 189,731, Sol 185,772. Output tokens per second of clock time (median): Opus 106.62, Sol 23.29.
- No cost comparison: cost was not measured the same way for both.

## Limits
- One run per job, small jobs, two languages, one machine, one day. We mostly use Claude, and a Claude Opus 5.5 session built the test kit; the judge was a third company and never saw names.

## Links
- [The results page](https://ccdocs.com/opus-versus-sol/)
- [The full write-up](https://ccdocs.com/opus-versus-sol/write-up/)
- [All 21 jobs, one page each](https://ccdocs.com/opus-versus-sol/#jobs)
- [t01 Kanban board web service](https://ccdocs.com/opus-versus-sol/t01/)
- [t02 Bug hunt](https://ccdocs.com/opus-versus-sol/t02/)
- [t03 Refactor](https://ccdocs.com/opus-versus-sol/t03/)
- [t04 Make it fast](https://ccdocs.com/opus-versus-sol/t04/)
- [t05 Vulnerable web service](https://ccdocs.com/opus-versus-sol/t05/)
- [t06 Database migration](https://ccdocs.com/opus-versus-sol/t06/)
- [t07 Race condition](https://ccdocs.com/opus-versus-sol/t07/)
- [t08 Write the tests](https://ccdocs.com/opus-versus-sol/t08/)
- [t09 Port Python to JavaScript](https://ccdocs.com/opus-versus-sol/t09/)
- [t10 Vague request](https://ccdocs.com/opus-versus-sol/t10/)
- [t11 Messy money](https://ccdocs.com/opus-versus-sol/t11/)
- [t12 Build from a spec](https://ccdocs.com/opus-versus-sol/t12/)
- [t13 Flaky web service client](https://ccdocs.com/opus-versus-sol/t13/)
- [t14 Big codebase questions](https://ccdocs.com/opus-versus-sol/t14/)
- [t15 Red build from logs](https://ccdocs.com/opus-versus-sol/t15/)
- [t16 Docker setup](https://ccdocs.com/opus-versus-sol/t16/)
- [t17 Fake-success trap](https://ccdocs.com/opus-versus-sol/t17/)
- [t18 Feature with changes](https://ccdocs.com/opus-versus-sol/t18/)
- [t19 Review code changes](https://ccdocs.com/opus-versus-sol/t19/)
- [t20 Issue to pull request](https://ccdocs.com/opus-versus-sol/t20/)
- [t21 Snake game](https://ccdocs.com/opus-versus-sol/t21/)
- [The raw data downloads](https://ccdocs.com/opus-versus-sol/#download)