Reducto · Document API
Reducto Extract on ResumeBench: 81.7 F1, rank 12 of 18
Reducto Extract ranks 12th of 18 on ResumeBench: 81.7 F1, $20 per 1,000 pages, 16 seconds per resume. It is 5.5 points behind LlamaParse Extract Agentic Plus. Qwen 3.8 Flash scores 84.1 for $2.98.
6th of 7 document APIs. Strongest on scanned & rotated (+7.0 points), rotated & warped pages (+6.3 points) and hidden text & prompt injections (+2.7 points). Weakest on handwritten (−12.1 points), three-column (−4.9 points) and busy page (−4.7 points).
- Rank
- 12/18
- F1
- 81.7
- Precision
- 83.7
- Recall
- 84.0
- Perfect resumes
- 16/300
- Per 1,000 pages
- $20
- Median per resume
- 16 s
Averaged over 300 resumes. Dates (scored separately): 96.8. 90% finish within 633 s. Cost from list price.
Where Reducto 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 | 39.4 | 51.5 | −12.1 | 18th |
| Image only | 28 | 72.0 | 71.4 | +0.6 | 9th |
| Sidebar or grid | 82 | 82.1 | 81.9 | +0.2 | 9th |
| Photo | 45 | 81.9 | 84.0 | −2.1 | 12th |
Distractors
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Rotated & warped pages | 21 | 89.2 | 82.9 | +6.3 | 6th |
| Hidden text & prompt injections | 14 | 87.2 | 84.5 | +2.7 | 8th |
| Content as images | 8 | 84.9 | 88.2 | −3.3 | 13th |
| Wording traps | 16 | 81.4 | 84.5 | −3.1 | 13th |
| Typos kept on purpose | 37 | 86.2 | 83.7 | +2.6 | 6th |
Extraction challenges
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Many entries | 83 | 79.5 | 81.9 | −2.4 | 13th |
| Long bullet lists | 101 | 80.2 | 81.4 | −1.2 | 12th |
| Split across pages | 85 | 79.7 | 82.2 | −2.5 | 12th |
| Big skills list | 60 | 83.2 | 83.6 | −0.4 | 11th |
| Custom sections | 105 | 78.2 | 80.8 | −2.6 | 12th |
| Combined headings | 37 | 80.2 | 82.7 | −2.5 | 12th |
| Promotions | 88 | 82.8 | 83.5 | −0.7 | 12th |
| Partial dates | 138 | 80.0 | 83.4 | −3.4 | 13th |
| Overlapping jobs | 69 | 80.4 | 81.7 | −1.4 | 12th |
| Accented text | 13 | 87.8 | 85.5 | +2.2 | 4th |
Noise
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Repeated header | 67 | 83.6 | 85.2 | −1.6 | 12th |
| References | 55 | 84.4 | 85.2 | −0.8 | 12th |
| Busy page | 18 | 75.4 | 80.1 | −4.7 | 13th |
Source
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Everyday resumes | 212 | 83.0 | 85.1 | −2.2 | 12th |
| Expert-built (Canva / LaTeX) | 32 | 80.0 | 83.6 | −3.6 | 13th |
| Resume-builder templates | 10 | 81.8 | 81.2 | +0.6 | 9th |
| Scanned & rotated | 19 | 89.9 | 82.9 | +7.0 | 6th |
| Distractor layouts | 18 | 82.3 | 85.3 | −3.1 | 13th |
Layout
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Single column | 195 | 81.3 | 84.0 | −2.6 | 13th |
| Two-column | 23 | 83.2 | 86.8 | −3.6 | 12th |
| Two-column sidebar | 54 | 81.6 | 80.0 | +1.6 | 8th |
| Three-column | 20 | 79.9 | 84.8 | −4.9 | 13th |
| Table grid | 8 | 90.8 | 94.1 | −3.3 | 14th |
Length
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Short | 148 | 85.3 | 87.1 | −1.8 | 11th |
| Medium | 90 | 82.5 | 84.7 | −2.2 | 12th |
| Long | 62 | 71.9 | 74.6 | −2.8 | 13th |
By section of the resume
F1 by resume section, against the median system.
| Section | Resumes | F1 | Median | Gap |
|---|---|---|---|---|
| Contact details | 300 | 93.4 | 93.5 | −0.1 |
| Experience | 300 | 84.0 | 86.3 | −2.3 |
| Education | 297 | 82.3 | 89.4 | −7.1 |
| Summary | 290 | 81.7 | 81.7 | +0.0 |
| Certifications | 244 | 81.3 | 81.1 | +0.2 |
| Awards | 102 | 73.5 | 75.4 | −1.9 |
| Skills | 290 | 93.0 | 92.1 | +0.9 |
| Languages | 169 | 84.3 | 84.3 | +0.0 |
| Custom sections | 216 | 44.4 | 49.9 | −5.5 |
| Interests | 64 | 92.8 | 91.7 | +1.1 |
| Volunteering | 80 | 82.7 | 84.5 | −1.8 |
| Profiles | 35 | 58.1 | 66.7 | −8.6 |
| Projects | 78 | 84.6 | 83.2 | +1.4 |
| Publications | 51 | 76.0 | 78.6 | −2.5 |
On the published resumes
Best and worst of the 20 public resumes.
| Resume | What makes it hard | F1 |
|---|---|---|
| Head of Developer Experience (Internal title: Jedi Master of Analytics) | Unusual job titles | 93.5 |
| Senior Manufacturing Engineer | Third-person prose, no bullets | 92.4 |
| Deputy Head of Mission | Rotated scan with hidden identity text | 89.3 |
| Senior Game Designer | Game designer's Canva resume | 61.7 |
| Senior ocularist and practice owner | Handwritten resume, six pages | 47.0 |
| senior Consultant - package Implementation | Handwritten resume | 40.0 |
How it was run
Reducto Extract API, default settings. Same PDF, schema and instructions as every other system, one run per resume (4 November 2026).
Reproduce it with the open-source harness (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 run reducto_extract --split test
$ resume-bench grade reducto_extract --split testClosest alternatives
Nearest in F1.
| Model | F1 | Per 1,000 pages | Median time | Head to head |
|---|---|---|---|---|
| Qwen 3.8 27B | 83.8 | $9.59 | 36 s | Reducto Extract vs Qwen 3.8 27B |
| LlamaParse Extract Cost-Effective | 83.5 | $8.18 | 29 s | Reducto Extract vs LlamaParse Extract Cost-Effective |
| GPT-5.5 | 83.5 | $87 | 48 s | Reducto Extract vs GPT-5.5 |
| Extend Extract | 81.0 | $12 | 31 s | Reducto Extract vs Extend Extract |
Reducto Extract against every other model
- Reducto Extract vs LlamaParse Extract Agentic Plus
- Reducto Extract vs LlamaParse Extract Agentic
- Reducto Extract vs Reducto Deep Extract
- Reducto Extract vs Extend Max Context
- Reducto Extract vs GPT-5.6
- Reducto Extract vs GPT-5.4 Nano
- Reducto Extract vs Claude Opus 4.8
- Reducto Extract vs Gemini 3.5 Flash
- Reducto Extract vs Gemma 4 26B
- Reducto Extract vs Qwen 3.8 Flash
- Reducto Extract vs Qwen 3.5 35B
- Reducto Extract vs Qwen 3.5 9B
- Reducto Extract vs GLM 5.3 Flash
Embed this result
Shows its current rank. Updates with each run.