Alibaba · Vision LLM
Qwen 3.8 Flash on ResumeBench: 84.1 F1, rank 8 of 18
Qwen 3.8 Flash ranks 8th of 18 on ResumeBench: 84.1 F1, $2.98 per 1,000 pages, 122 seconds per resume. It is 3.1 points behind LlamaParse Extract Agentic Plus.
4th of 11 vision LLMs. Strongest on hidden text & prompt injections (+4.5 points), scanned & rotated (+3.4 points) and image only (+3.3 points). Weakest on overlapping jobs (−0.3 points), busy page (−0.3 points) and distractor layouts (−0.2 points).
- Rank
- 8/18
- F1
- 84.1
- Precision
- 85.4
- Recall
- 86.2
- Perfect resumes
- 25/300
- Per 1,000 pages
- $2.98
- Median per resume
- 122 s
Averaged over 300 resumes. Dates (scored separately): 76.5. 90% finish within 246 s.
Where Qwen 3.8 Flash 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 | 53.5 | 51.5 | +2.0 | 5th |
| Image only | 28 | 74.7 | 71.4 | +3.3 | 8th |
| Sidebar or grid | 82 | 83.8 | 81.9 | +1.9 | 8th |
| Photo | 45 | 86.3 | 84.0 | +2.3 | 3rd |
Distractors
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Rotated & warped pages | 21 | 86.0 | 82.9 | +3.1 | 9th |
| Hidden text & prompt injections | 14 | 89.0 | 84.5 | +4.5 | 7th |
| Content as images | 8 | 88.3 | 88.2 | +0.0 | 8th |
| Wording traps | 16 | 86.0 | 84.5 | +1.5 | 5th |
| Typos kept on purpose | 37 | 85.7 | 83.7 | +2.1 | 7th |
Extraction challenges
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Many entries | 83 | 82.1 | 81.9 | +0.2 | 9th |
| Long bullet lists | 101 | 82.1 | 81.4 | +0.6 | 9th |
| Split across pages | 85 | 82.6 | 82.2 | +0.4 | 9th |
| Big skills list | 60 | 83.5 | 83.6 | −0.1 | 10th |
| Custom sections | 105 | 80.8 | 80.8 | +0.0 | 9th |
| Combined headings | 37 | 82.6 | 82.7 | −0.0 | 10th |
| Promotions | 88 | 83.8 | 83.5 | +0.3 | 9th |
| Partial dates | 138 | 83.3 | 83.4 | −0.1 | 10th |
| Overlapping jobs | 69 | 81.4 | 81.7 | −0.3 | 10th |
| Accented text | 13 | 86.9 | 85.5 | +1.4 | 6th |
Noise
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Repeated header | 67 | 85.1 | 85.2 | −0.1 | 10th |
| References | 55 | 85.4 | 85.2 | +0.2 | 9th |
| Busy page | 18 | 79.8 | 80.1 | −0.3 | 10th |
Source
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Everyday resumes | 212 | 85.1 | 85.1 | −0.1 | 10th |
| Expert-built (Canva / LaTeX) | 32 | 84.4 | 83.6 | +0.7 | 7th |
| Resume-builder templates | 10 | 84.0 | 81.2 | +2.9 | 5th |
| Scanned & rotated | 19 | 86.3 | 82.9 | +3.4 | 9th |
| Distractor layouts | 18 | 85.1 | 85.3 | −0.2 | 11th |
Layout
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Single column | 195 | 83.8 | 84.0 | −0.1 | 10th |
| Two-column | 23 | 87.1 | 86.8 | +0.3 | 8th |
| Two-column sidebar | 54 | 81.6 | 80.0 | +1.6 | 9th |
| Three-column | 20 | 84.9 | 84.8 | +0.1 | 9th |
| Table grid | 8 | 95.5 | 94.1 | +1.4 | 4th |
Length
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Short | 148 | 87.2 | 87.1 | +0.1 | 8th |
| Medium | 90 | 85.3 | 84.7 | +0.6 | 6th |
| Long | 62 | 75.0 | 74.6 | +0.3 | 9th |
By section of the resume
F1 by resume section, against the median system.
| Section | Resumes | F1 | Median | Gap |
|---|---|---|---|---|
| Contact details | 298 | 93.6 | 93.5 | +0.1 |
| Experience | 298 | 88.0 | 86.3 | +1.7 |
| Education | 295 | 90.1 | 89.4 | +0.7 |
| Summary | 288 | 81.3 | 81.7 | −0.5 |
| Certifications | 248 | 82.2 | 81.1 | +1.1 |
| Awards | 94 | 75.5 | 75.4 | +0.1 |
| Skills | 285 | 90.0 | 92.1 | −2.1 |
| Languages | 167 | 84.6 | 84.3 | +0.3 |
| Custom sections | 210 | 53.9 | 49.9 | +4.0 |
| Interests | 63 | 91.1 | 91.7 | −0.6 |
| Volunteering | 78 | 83.9 | 84.5 | −0.6 |
| Profiles | 35 | 73.3 | 66.7 | +6.6 |
| Projects | 79 | 83.7 | 83.2 | +0.5 |
| Publications | 51 | 81.8 | 78.6 | +3.3 |
On the published resumes
Best and worst of the 20 public resumes.
| Resume | What makes it hard | F1 |
|---|---|---|
| Head of Events & Public Space Operations | Table grid with emoji bullets | 96.3 |
| Senior Customer Service Representative | Busy Canva page | 94.9 |
| Head of Developer Experience (Internal title: Jedi Master of Analytics) | Unusual job titles | 88.8 |
| Senior Game Designer | Game designer's Canva resume | 67.7 |
| senior Consultant - package Implementation | Handwritten resume | 44.9 |
| Senior ocularist and practice owner | Handwritten resume, six pages | 40.2 |
How it was run
Alibaba Cloud Model Studio, qwen3.8-flash, 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 Flash vs Extend Max Context |
| Qwen 3.8 27B | 83.8 | $9.59 | 36 s | Qwen 3.8 Flash vs Qwen 3.8 27B |
| LlamaParse Extract Cost-Effective | 83.5 | $8.18 | 29 s | Qwen 3.8 Flash vs LlamaParse Extract Cost-Effective |
| GPT-5.5 | 83.5 | $87 | 48 s | Qwen 3.8 Flash vs GPT-5.5 |
Qwen 3.8 Flash against every other model
- Qwen 3.8 Flash vs LlamaParse Extract Agentic Plus
- Qwen 3.8 Flash vs LlamaParse Extract Agentic
- Qwen 3.8 Flash vs Reducto Deep Extract
- Qwen 3.8 Flash vs Reducto Extract
- Qwen 3.8 Flash vs Extend Extract
- Qwen 3.8 Flash vs GPT-5.6
- Qwen 3.8 Flash vs GPT-5.4 Nano
- Qwen 3.8 Flash vs Claude Opus 4.8
- Qwen 3.8 Flash vs Gemini 3.5 Flash
- Qwen 3.8 Flash vs Gemma 4 26B
- Qwen 3.8 Flash vs Qwen 3.5 35B
- Qwen 3.8 Flash vs Qwen 3.5 9B
- Qwen 3.8 Flash vs GLM 5.3 Flash
Embed this result
Shows its current rank. Updates with each run.