No pick · 10 Nov
No model reached 90% right.
Highest , below 90%: GPT-6 Luna
650 right answers per $1
$1.52 per 1,000 right
Buy at OpenAI (opens OpenAI, another site)Model comparison
How the models compare
Documents · 10 Nov
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
  GPT-6 Luna | 87 | $1.52 |
  NL;NL DeskNot open yet. | 88 | $1.76 |
  Claude Haiku 5.5 | 84 | $1.78 |
 GLM-5.3 Flash | 82 | $2.45 |
  Gemini 3.1 Flash Lite | 78 | $5.73 |
| NL;NL Pull 4B | 24 | $17.66 |
| NL;NL Pull 12B | 8 | $52.48 |
Earlier results
Documents · 9 Nov
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| NL;NL DeskNot open yet. | 91 | $1.71 |
| GPT-6 Luna | 87 | $1.52 |
| Claude Haiku 5.5 | 84 | $1.78 |
| GLM-5.3 Flash | 84 | $2.37 |
| Gemini 3.1 Flash Lite | 81 | $5.61 |
| NL;NL Pull 4B | 26 | $9.43 |
| NL;NL Pull 12B | 8 | $36.38 |
Documents · 8 Nov
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| GPT-6 LunaPick | 91 | $1.47 |
| Claude Haiku 5.5 | 86 | $1.75 |
| NL;NL DeskNot open yet. | 86 | $1.81 |
| GLM-5.3 Flash | 78 | $2.56 |
| Gemini 3.1 Flash Lite | 82 | $5.58 |
| NL;NL Pull 4B | 25 | $20.22 |
| NL;NL Pull 12B | 14 | $27.86 |
Documents · 7 Nov
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| GPT-6 Luna | 87 | $1.53 |
| NL;NL DeskNot open yet. | 89 | $1.75 |
| Claude Haiku 5.5 | 86 | $1.75 |
| GLM-5.3 Flash | 85 | $2.34 |
| Gemini 3.1 Flash Lite | 81 | $5.75 |
| NL;NL Pull 4B | 33 | $18.54 |
| NL;NL Pull 12B | 10 | $42.75 |
Documents · 6 Nov
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| GPT-6 Luna | 87 | $1.53 |
| Claude Haiku 5.5 | 86 | $1.74 |
| NL;NL DeskNot open yet. | 86 | $1.80 |
| GLM-5.3 Flash | 79 | $2.54 |
| Gemini 3.1 Flash Lite | 87 | $5.26 |
| NL;NL Pull 4B | 26 | $9.52 |
| NL;NL Pull 12B | 12 | $26.82 |
Documents · 5 Nov
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| GPT-6 Luna | 89 | $1.49 |
| Claude Haiku 5.5 | 86 | $1.74 |
| NL;NL DeskNot open yet. | 88 | $1.77 |
| GLM-5.3 Flash | 80 | $2.50 |
| Gemini 3.1 Flash Lite | 83 | $5.46 |
| NL;NL Pull 4B | 19 | $27.95 |
| NL;NL Pull 12B | 12 | $33.61 |
Documents · 4 Nov
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| NL;NL DeskNot open yet. | 90 | $1.73 |
| 89 | $1.50 | |
Invented 3Missed 0Unreadable 0 | ||
| 86 | $1.74 | |
Invented 2Missed 1Unreadable 2 | ||
| 83 | $2.43 | |
Invented 3Missed 1Unreadable 0 | ||
| 80 | $5.66 | |
Invented 3Missed 1Unreadable 0 | ||
| NL;NL Pull 4B | 22 | $23.75 |
| NL;NL Pull 12B | 12 | $38.86 |
Text · 28 Oct
| Model | Right (%) | Cost per 1,000 right |
|---|---|---|
| GPT-6 LunaPick | 100 | $0.12 |
| Claude Haiku 5.5 | 98 | $0.18 |
| GLM-5.3 Flash | 98 | $0.27 |
The test
- Goes in
- Full documents; text items in the earlier test
- Comes out
- The named fields
- Right means
- Every tested field matches the answer key
- Checked by
- A script, against the answer key.
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.