Science & Tech · AI · 4 hrs ago
Claude Code, Codex and Cursor all passed the same coding test
A developer tested Claude Code, Codex and Cursor on the same coding task.
The three tools all used Claude Sonnet 5.5 with high reasoning.
The task was to fix four bugs and add features to a Kindle dashboard linked to Home Assistant, within 30 minutes and a $10 budget for each tool.
The Kindle’s old software requires JavaScript without modern features, adding a constraint to the work.
All three tools scored 16 out of 16 on the final code, though their processes differed.
The test also required different ways to connect the model to each tool, so the setup was not identical in every respect.
The results suggest the coding tool can matter beyond the model, but no further test outcome or next step is given.
XDA tested Claude Code, Codex and Cursor on the same coding task using Claude Sonnet 5.5.
Each tool received the same buggy WebSocket proxy project, 30 minutes and a $10 budget.
All three produced final code that scored 16 out of 16 on the test.
The tools differed in how they used the model and in the setup needed to run it.
The project involved a Kindle dashboard for Home Assistant and a proxy server.
- Who
- XDA writer Joe tested Claude Code, Codex and Cursor.
- What
- The tools worked on the same coding task using Claude Sonnet 5.5; each final code submission scored 16 out of 16.
- When
- The article was published on Oct. 22, 2026. The test date is not stated.
- Where
- Not stated.
- Why
- The writer wanted to test whether the coding tool affected results when the model and task were kept the same.
This story does not have two clearly opposing sides.
The writer prepared a Kindle dashboard project with four bugs and a request for an auto-reconnect feature.
Claude Code, Codex and Cursor each received the same model, project, 30 minutes and $10.
All three tools' final code scored 16 out of 16.
- Tools tested
- Claude Code, Codex and Cursor
- Model
- Claude Sonnet 5.5 at high reasoning
- Time allowed
- 30 minutes per tool
- Budget
- $10 per tool
- Final score
- 16 out of 16 for each tool
- Project
- Kindle dashboard for Home Assistant and a WebSocket proxy
What people are saying
Every one of them scored a perfect 16 out of 16 on the final code, but that's the last thing they had in common.