Alibaba · Vision LLM
Qwen 3.8 27B on ResumeBench: 83.8 F1, rank 9 of 18
Qwen 3.8 27B ranks 9th of 18 on ResumeBench: 83.8 F1, $9.59 per 1,000 pages, 36 seconds per resume. It is 3.4 points behind LlamaParse Extract Agentic Plus. Qwen 3.8 Flash scores 84.1 for $2.98.
5th of 11 vision LLMs. Strongest on resume-builder templates (+5.5 points), wording traps (+3.1 points) and distractor layouts (+2.3 points). Weakest on scanned & rotated (−4.5 points), rotated & warped pages (−4.2 points) and hidden text & prompt injections (−2.2 points).
- Rank
- 9/18
- F1
- 83.8
- Precision
- 85.4
- Recall
- 85.5
- Perfect resumes
- 31/300
- Per 1,000 pages
- $9.59
- Median per resume
- 36 s
Averaged over 300 resumes. Dates (scored separately): 91.6. 90% finish within 167 s.
Where Qwen 3.8 27B 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 | 51.6 | 51.5 | +0.1 | 9th |
| Image only | 28 | 70.5 | 71.4 | −0.8 | 11th |
| Sidebar or grid | 82 | 81.8 | 81.9 | −0.2 | 10th |
| Photo | 45 | 84.1 | 84.0 | +0.0 | 9th |
Distractors
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Rotated & warped pages | 21 | 78.7 | 82.9 | −4.2 | 11th |
| Hidden text & prompt injections | 14 | 82.4 | 84.5 | −2.2 | 11th |
| Content as images | 8 | 89.5 | 88.2 | +1.3 | 6th |
| Wording traps | 16 | 87.6 | 84.5 | +3.1 | 4th |
| Typos kept on purpose | 37 | 82.9 | 83.7 | −0.7 | 10th |
Extraction challenges
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Many entries | 83 | 81.6 | 81.9 | −0.2 | 10th |
| Long bullet lists | 101 | 80.8 | 81.4 | −0.6 | 10th |
| Split across pages | 85 | 81.8 | 82.2 | −0.4 | 10th |
| Big skills list | 60 | 82.9 | 83.6 | −0.8 | 12th |
| Custom sections | 105 | 81.0 | 80.8 | +0.2 | 8th |
| Combined headings | 37 | 83.7 | 82.7 | +1.0 | 5th |
| Promotions | 88 | 83.2 | 83.5 | −0.3 | 10th |
| Partial dates | 138 | 83.6 | 83.4 | +0.1 | 9th |
| Overlapping jobs | 69 | 82.4 | 81.7 | +0.6 | 8th |
| Accented text | 13 | 84.9 | 85.5 | −0.6 | 11th |
Noise
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Repeated header | 67 | 85.2 | 85.2 | +0.1 | 9th |
| References | 55 | 84.9 | 85.2 | −0.2 | 10th |
| Busy page | 18 | 80.4 | 80.1 | +0.3 | 9th |
Source
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Everyday resumes | 212 | 85.2 | 85.1 | +0.1 | 9th |
| Expert-built (Canva / LaTeX) | 32 | 83.4 | 83.6 | −0.2 | 11th |
| Resume-builder templates | 10 | 86.7 | 81.2 | +5.5 | 1st |
| Scanned & rotated | 19 | 78.4 | 82.9 | −4.5 | 11th |
| Distractor layouts | 18 | 87.6 | 85.3 | +2.3 | 4th |
Layout
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Single column | 195 | 84.1 | 84.0 | +0.1 | 9th |
| Two-column | 23 | 87.9 | 86.8 | +1.1 | 7th |
| Two-column sidebar | 54 | 78.4 | 80.0 | −1.6 | 10th |
| Three-column | 20 | 85.5 | 84.8 | +0.7 | 7th |
| Table grid | 8 | 95.7 | 94.1 | +1.6 | 3rd |
Length
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Short | 148 | 87.1 | 87.1 | −0.0 | 10th |
| Medium | 90 | 85.5 | 84.7 | +0.8 | 5th |
| Long | 62 | 73.4 | 74.6 | −1.2 | 12th |
By section of the resume
F1 by resume section, against the median system.
| Section | Resumes | F1 | Median | Gap |
|---|---|---|---|---|
| Contact details | 300 | 93.2 | 93.5 | −0.4 |
| Experience | 300 | 86.2 | 86.3 | −0.2 |
| Education | 297 | 87.2 | 89.4 | −2.3 |
| Summary | 290 | 81.7 | 81.7 | +0.0 |
| Certifications | 244 | 82.7 | 81.1 | +1.6 |
| Awards | 97 | 78.4 | 75.4 | +3.0 |
| Skills | 284 | 92.7 | 92.1 | +0.6 |
| Languages | 168 | 85.4 | 84.3 | +1.0 |
| Custom sections | 216 | 51.2 | 49.9 | +1.3 |
| Interests | 63 | 94.6 | 91.7 | +2.9 |
| Volunteering | 79 | 84.7 | 84.5 | +0.1 |
| Profiles | 34 | 72.5 | 66.7 | +5.8 |
| Projects | 78 | 85.5 | 83.2 | +2.3 |
| Publications | 51 | 77.2 | 78.6 | −1.3 |
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 | 96.3 |
| Supervisory Human Resources Specialist, GS-0201-15 | Resume-builder template | 88.5 |
| Head of Events & Public Space Operations | Table grid with emoji bullets | 88.1 |
| Deputy Head of Mission | Rotated scan with hidden identity text | 65.6 |
| senior Consultant - package Implementation | Handwritten resume | 46.5 |
| Senior ocularist and practice owner | Handwritten resume, six pages | 41.2 |
How it was run
Qwen3.8-27B (open weights), page images in, JSON schema out. 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 |
|---|---|---|---|---|
| Extend Max Context | 84.7 | $50 | 43 s | Qwen 3.8 27B vs Extend Max Context |
| Qwen 3.8 Flash | 84.1 | $2.98 | 122 s | Qwen 3.8 27B vs Qwen 3.8 Flash |
| LlamaParse Extract Cost-Effective | 83.5 | $8.18 | 29 s | Qwen 3.8 27B vs LlamaParse Extract Cost-Effective |
| GPT-5.5 | 83.5 | $87 | 48 s | Qwen 3.8 27B vs GPT-5.5 |
Qwen 3.8 27B against every other model
- Qwen 3.8 27B vs LlamaParse Extract Agentic Plus
- Qwen 3.8 27B vs LlamaParse Extract Agentic
- Qwen 3.8 27B vs Reducto Deep Extract
- Qwen 3.8 27B vs Reducto Extract
- Qwen 3.8 27B vs Extend Extract
- Qwen 3.8 27B vs GPT-5.6
- Qwen 3.8 27B vs GPT-5.4 Nano
- Qwen 3.8 27B vs Claude Opus 4.8
- Qwen 3.8 27B vs Gemini 3.5 Flash
- Qwen 3.8 27B vs Gemma 4 26B
- Qwen 3.8 27B vs Qwen 3.5 35B
- Qwen 3.8 27B vs Qwen 3.5 9B
- Qwen 3.8 27B vs GLM 5.3 Flash
Embed this result
Shows its current rank. Updates with each run.