Which resume extraction model handles rotated and warped scans best?
21 of the 300 resumes are rotated and warped scans. The typical system scores 0.4 points higher on them. Best: LlamaParse Extract Agentic Plus, 92.5 F1. LlamaParse Extract Agentic is within two points for $23 per 1,000 pages.
Image-only scans with pages at different rotations, warped pages and photos of printouts. Best document API: LlamaParse Extract Agentic Plus (92.5). Best vision LLM: Gemini 3.5 Flash (87.4). LlamaParse Extract Cost-Effective ranks 6 places higher here than overall.
Every system on rotated and warped scans
F1 on the 21 rotated and warped scans, against each system’s F1 on all 300 resumes. Cost and time are for all resumes.
| Model | F1 here | F1 overall | Gap | Per 1,000 pages | Median time |
|---|---|---|---|---|---|
| LlamaParse Extract Agentic Plus | 92.5 | 87.2 | +5.3 | $54 | 70 s |
| LlamaParse Extract Agentic | 90.9 | 85.5 | +5.4 | $23 | 55 s |
| Extend Max Context | 90.8 | 84.7 | +6.0 | $50 | 43 s |
| LlamaParse Extract Cost-Effective | 89.6 | 83.5 | +6.1 | $8.18 | 29 s |
| Reducto Deep Extract | 89.3 | 85.6 | +3.7 | $40 | 65 s |
| Reducto Extract | 89.2 | 81.7 | +7.5 | $20 | 16 s |
| Gemini 3.5 Flash | 87.4 | 85.5 | +2.0 | $35 | 43 s |
| Claude Opus 4.8 | 87.4 | 85.5 | +1.9 | $90 | 31 s |
| Qwen 3.8 Flash | 86.0 | 84.1 | +1.9 | $2.98 | 122 s |
| Extend Extract | 79.8 | 81.0 | −1.1 | $12 | 31 s |
| Qwen 3.8 27B | 78.7 | 83.8 | −5.0 | $9.59 | 36 s |
| Qwen 3.5 35B | 76.5 | 77.9 | −1.4 | $3.91 | 78 s |
| GPT-5.6 | 73.2 | 84.7 | −11.5 | $74 | 38 s |
| Qwen 3.5 9B | 66.5 | 76.0 | −9.6 | $1.31 | 138 s |
| GPT-5.5 | 63.4 | 83.5 | −20.1 | $87 | 48 s |
| GLM 5.3 Flash | 50.3 | 59.2 | −9.0 | $0.80 | 46 s |
| Gemma 4 26B | 43.4 | 62.8 | −19.4 | $0.56 | 29 s |
| GPT-5.4 Nano | 11.7 | 61.9 | −50.2 | $2.76 | 20 s |
What to do about it
If you get phone photos or bad scans, rank parsers on this table.
Published examples
2 public resumes of this kind.