Idle tomato tycoon: 582 lines vs 2,530
ChatGPT vs Claude
We test AI so you know what actually works.
Real-world tests, practical guides and side-by-side model comparisons — for people who actually use this stuff, not for people reading about it.
Just published
Which AI are you?
Playground8 questions · about 1 minute
One holographic tree, six reasoning modes
Gemini vs DeepSeek
Gemini introduces itself, four briefs from bare to dictated
Gemini · 4 attempts
Gemini Flash vs Pro: a light-spirit arcade game in one file
Gemini 3.6 Flash (High) vs Gemini 3.1 Pro (High)
Endless racer: eleven minutes against two
Gemini 3.5 Flash (High) vs Gemini 3.1 Pro (High)
Live scoreboard
Who has actually been tested
Every model that has sat a published test, and how long it took over them.
- Gemini55m 0s average
- ChatGPT47m 29s average
- Claude229m 40s average
- DeepSeek1time not recorded
Nothing has been called yet — every round is still open, and the bars count who turned up rather than who won.
Popular right now
By what people opened this week
What are you trying to do with AI?
Compare modelsThe same brief, run on each model, with the outputs published untouched.8 testsBuild & codeClaude Code, web work, debugging, agents.Being writtenUse MCPServers, integrations, workflows, security.1 write-upAutomate workBrowsers, research, data, the repetitive parts.Being writtenWork with dataPDFs, CSVs, SQL, spreadsheets.Being writtenFix somethingSymptoms, in the words people actually use.Being written
What are you trying to do with AI?
AI for developers
Claude Code, MCP and AI-assisted coding — the parts that only start to matter once you are past the demo.
The first technical write-ups are being tested before they are written.
Something not working?
Symptoms in the words people actually use, and the fixes that were tried rather than the ones that sound plausible.
No fixes published yet. This is where they will live.
MCP, without the fluff
Which server to reach for, where it actually helps, and how it behaves in a real workflow.
MCPMeasured6 min
stdio vs HTTP: what an MCP transport actually costs
Practical guides
Nothing here yet — the first ones are being run.
We publish the prompts, the outputs and the measurements.
- Published tests
- 8
- Model runs
- 23
- Automated checks
- 91
- Community votes
- 1
- The same prompt, word for word.
- Outputs published unedited.
- Real completion times, measured in the app.
- A method you can read and argue with.