Pick · 10 Nov
Highest (right answers per dollar) among models at least 90% right; no pick if none qualify.
GPT-6 Luna
2,300 right answers per $1
$0.43 per 1,000 right
Buy at OpenAI (opens OpenAI, another site)Model comparison
How the models compare
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| GPT-6 LunaPick | 100 | $0.43 |
  Claude Haiku 5.5 | 98 | $0.81 |
 GLM-5.3 Flash | 96 | $4.31 |
  Gemini 3.1 Flash Lite | 84 | $1.22 |
The test
- Goes in
- A broken Python function
- Comes out
- The repaired function
- Right means
- It passes every hidden test case
- Checked by
- A program that runs the tests
Desk · Not open yet.
If another company’s model has higher than Desk that day, we’ll link you to the highest- model before you send a job. You will still be able to use Desk.