Reducto · Document API
Reducto Deep Extract on ResumeBench: 85.6 F1, rank 2 of 18
Reducto Deep Extract ranks 2nd of 18 on ResumeBench: 85.6 F1, $40 per 1,000 pages, 65 seconds per resume. It is 1.6 points behind LlamaParse Extract Agentic Plus.
2nd of 7 document APIs. Strongest on scanned & rotated (+7.2 points), rotated & warped pages (+6.3 points) and hidden text & prompt injections (+5.5 points). Weakest on resume-builder templates (−3.6 points), handwritten (−2.1 points) and table grid (−0.9 points).
- Rank
- 2/18
- F1
- 85.6
- Precision
- 87.5
- Recall
- 87.3
- Perfect resumes
- 36/300
- Per 1,000 pages
- $40
- Median per resume
- 65 s
Averaged over 300 resumes. Dates (scored separately): 97.7. 90% finish within 92 s. Cost from list price.
Where Reducto Deep Extract wins and loses
F1 by kind of resume, against the median system.
Reading the page
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Handwritten | 9 | 49.4 | 51.5 | −2.1 | 16th |
| Image only | 28 | 76.4 | 71.4 | +5.1 | 5th |
| Sidebar or grid | 82 | 85.2 | 81.9 | +3.3 | 3rd |
| Photo | 45 | 86.4 | 84.0 | +2.3 | 2nd |
Distractors
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Rotated & warped pages | 21 | 89.3 | 82.9 | +6.3 | 5th |
| Hidden text & prompt injections | 14 | 90.1 | 84.5 | +5.5 | 5th |
| Content as images | 8 | 88.0 | 88.2 | −0.2 | 11th |
| Wording traps | 16 | 84.2 | 84.5 | −0.3 | 10th |
| Typos kept on purpose | 37 | 87.9 | 83.7 | +4.2 | 5th |
Extraction challenges
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Many entries | 83 | 84.3 | 81.9 | +2.5 | 4th |
| Long bullet lists | 101 | 84.5 | 81.4 | +3.1 | 4th |
| Split across pages | 85 | 84.1 | 82.2 | +1.9 | 6th |
| Big skills list | 60 | 85.8 | 83.6 | +2.2 | 4th |
| Custom sections | 105 | 82.6 | 80.8 | +1.7 | 4th |
| Combined headings | 37 | 82.7 | 82.7 | +0.0 | 9th |
| Promotions | 88 | 86.0 | 83.5 | +2.5 | 3rd |
| Partial dates | 138 | 85.2 | 83.4 | +1.7 | 4th |
| Overlapping jobs | 69 | 85.9 | 81.7 | +4.1 | 2nd |
| Accented text | 13 | 85.5 | 85.5 | −0.1 | 10th |
Noise
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Repeated header | 67 | 87.3 | 85.2 | +2.1 | 3rd |
| References | 55 | 90.2 | 85.2 | +5.0 | 3rd |
| Busy page | 18 | 83.6 | 80.1 | +3.5 | 2nd |
Source
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Everyday resumes | 212 | 87.2 | 85.1 | +2.1 | 3rd |
| Expert-built (Canva / LaTeX) | 32 | 84.6 | 83.6 | +1.0 | 6th |
| Resume-builder templates | 10 | 77.6 | 81.2 | −3.6 | 15th |
| Scanned & rotated | 19 | 90.1 | 82.9 | +7.2 | 5th |
| Distractor layouts | 18 | 85.6 | 85.3 | +0.2 | 8th |
Layout
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Single column | 195 | 85.5 | 84.0 | +1.6 | 4th |
| Two-column | 23 | 86.9 | 86.8 | +0.1 | 9th |
| Two-column sidebar | 54 | 83.8 | 80.0 | +3.8 | 3rd |
| Three-column | 20 | 86.0 | 84.8 | +1.2 | 3rd |
| Table grid | 8 | 93.2 | 94.1 | −0.9 | 12th |
Length
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Short | 148 | 88.7 | 87.1 | +1.6 | 3rd |
| Medium | 90 | 85.9 | 84.7 | +1.2 | 2nd |
| Long | 62 | 77.5 | 74.6 | +2.8 | 6th |
By section of the resume
F1 by resume section, against the median system.
| Section | Resumes | F1 | Median | Gap |
|---|---|---|---|---|
| Contact details | 300 | 94.2 | 93.5 | +0.7 |
| Experience | 300 | 90.6 | 86.3 | +4.3 |
| Education | 297 | 92.0 | 89.4 | +2.6 |
| Summary | 290 | 85.2 | 81.7 | +3.4 |
| Certifications | 252 | 81.4 | 81.1 | +0.3 |
| Awards | 94 | 79.4 | 75.4 | +4.0 |
| Skills | 285 | 93.9 | 92.1 | +1.8 |
| Languages | 167 | 81.8 | 84.3 | −2.6 |
| Custom sections | 212 | 51.3 | 49.9 | +1.4 |
| Interests | 64 | 93.8 | 91.7 | +2.0 |
| Volunteering | 77 | 89.5 | 84.5 | +5.0 |
| Profiles | 34 | 69.6 | 66.7 | +2.9 |
| Projects | 77 | 88.1 | 83.2 | +4.9 |
| Publications | 50 | 79.3 | 78.6 | +0.8 |
On the published resumes
Best and worst of the 20 public resumes.
| Resume | What makes it hard | F1 |
|---|---|---|
| Senior Manufacturing Engineer | Third-person prose, no bullets | 93.9 |
| Deputy Head of Mission | Rotated scan with hidden identity text | 89.2 |
| Lead Respiratory Therapist (Adult ICU) | Long resume, 34 entries | 88.9 |
| senior Consultant - package Implementation | Handwritten resume | 54.8 |
| Senior Game Designer | Game designer's Canva resume | 52.9 |
| Senior ocularist and practice owner | Handwritten resume, six pages | 37.4 |
How it was run
Reducto Extract API with Deep Extract on. Same PDF, schema and instructions as every other system, one run per resume (4 November 2026).
Run the open-source harness: list its pipelines, or grade your own predictions (GitHub):
$ pip install "resume-bench[runners] @ git+https://github.com/careerflow/ResumeBench"
$ hf auth login # the dataset is gated: accept its terms on Hugging Face first
$ resume-bench download
$ resume-bench providers
$ resume-bench grade-file my_predictions.jsonl --split testClosest alternatives
Nearest in F1.
| Model | F1 | Per 1,000 pages | Median time | Head to head |
|---|---|---|---|---|
| LlamaParse Extract Agentic | 85.5 | $23 | 55 s | Reducto Deep Extract vs LlamaParse Extract Agentic |
| Claude Opus 4.8 | 85.5 | $90 | 31 s | Reducto Deep Extract vs Claude Opus 4.8 |
| Gemini 3.5 Flash | 85.5 | $35 | 43 s | Reducto Deep Extract vs Gemini 3.5 Flash |
| GPT-5.6 | 84.7 | $74 | 38 s | Reducto Deep Extract vs GPT-5.6 |
Reducto Deep Extract against every other model
- Reducto Deep Extract vs LlamaParse Extract Agentic Plus
- Reducto Deep Extract vs LlamaParse Extract Cost-Effective
- Reducto Deep Extract vs Reducto Extract
- Reducto Deep Extract vs Extend Max Context
- Reducto Deep Extract vs Extend Extract
- Reducto Deep Extract vs GPT-5.5
- Reducto Deep Extract vs GPT-5.4 Nano
- Reducto Deep Extract vs Gemma 4 26B
- Reducto Deep Extract vs Qwen 3.8 Flash
- Reducto Deep Extract vs Qwen 3.8 27B
- Reducto Deep Extract vs Qwen 3.5 35B
- Reducto Deep Extract vs Qwen 3.5 9B
- Reducto Deep Extract vs GLM 5.3 Flash
Embed this result
Shows its current rank. Updates with each run.