Zhipu AI · Vision LLM
GLM 5.3 Flash on ResumeBench: 59.2 F1, rank 18 of 18
GLM 5.3 Flash ranks 18th of 18 on ResumeBench: 59.2 F1, $0.80 per 1,000 pages, 46 seconds per resume. It is 27.9 points behind LlamaParse Extract Agentic Plus. Gemma 4 26B scores 62.8 for $0.56.
11th of 11 vision LLMs. Strongest on handwritten (+3.1 points), short (−7.2 points) and table grid (−11.1 points). Weakest on content as images (−77.8 points), distractor layouts (−67.5 points) and wording traps (−64.3 points).
- Rank
- 18/18
- F1
- 59.2
- Precision
- 71.8
- Recall
- 62.0
- Perfect resumes
- 23/300
- Per 1,000 pages
- $0.80
- Median per resume
- 46 s
Averaged over 300 resumes. Dates (scored separately): 63.6. 90% finish within 145 s.
Where GLM 5.3 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 | 54.6 | 51.5 | +3.1 | 4th |
| Image only | 28 | 53.3 | 71.4 | −18.1 | 16th |
| Sidebar or grid | 82 | 63.5 | 81.9 | −18.4 | 16th |
| Photo | 45 | 64.0 | 84.0 | −20.1 | 17th |
Distractors
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Rotated & warped pages | 21 | 50.3 | 82.9 | −32.7 | 16th |
| Hidden text & prompt injections | 14 | 46.4 | 84.5 | −38.1 | 17th |
| Content as images | 8 | 10.4 | 88.2 | −77.8 | 18th |
| Wording traps | 16 | 20.2 | 84.5 | −64.3 | 18th |
| Typos kept on purpose | 37 | 36.3 | 83.7 | −47.3 | 17th |
Extraction challenges
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Many entries | 83 | 43.1 | 81.9 | −38.8 | 18th |
| Long bullet lists | 101 | 42.8 | 81.4 | −38.7 | 18th |
| Split across pages | 85 | 51.2 | 82.2 | −31.0 | 18th |
| Big skills list | 60 | 64.0 | 83.6 | −19.7 | 17th |
| Custom sections | 105 | 48.2 | 80.8 | −32.6 | 18th |
| Combined headings | 37 | 65.7 | 82.7 | −17.0 | 16th |
| Promotions | 88 | 59.2 | 83.5 | −24.3 | 17th |
| Partial dates | 138 | 52.2 | 83.4 | −31.3 | 18th |
| Overlapping jobs | 69 | 50.4 | 81.7 | −31.3 | 18th |
| Accented text | 13 | 51.1 | 85.5 | −34.4 | 18th |
Noise
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Repeated header | 67 | 66.1 | 85.2 | −19.0 | 16th |
| References | 55 | 52.1 | 85.2 | −33.1 | 18th |
| Busy page | 18 | 48.2 | 80.1 | −31.9 | 18th |
Source
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Everyday resumes | 212 | 66.3 | 85.1 | −18.8 | 17th |
| Expert-built (Canva / LaTeX) | 32 | 42.9 | 83.6 | −40.8 | 18th |
| Resume-builder templates | 10 | 48.8 | 81.2 | −32.3 | 18th |
| Scanned & rotated | 19 | 54.8 | 82.9 | −28.1 | 16th |
| Distractor layouts | 18 | 17.8 | 85.3 | −67.5 | 18th |
Layout
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Single column | 195 | 57.0 | 84.0 | −26.9 | 18th |
| Two-column | 23 | 62.6 | 86.8 | −24.3 | 18th |
| Two-column sidebar | 54 | 61.2 | 80.0 | −18.8 | 16th |
| Three-column | 20 | 62.0 | 84.8 | −22.8 | 18th |
| Table grid | 8 | 83.0 | 94.1 | −11.1 | 16th |
Length
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Short | 148 | 79.9 | 87.1 | −7.2 | 14th |
| Medium | 90 | 24.2 | 84.7 | −60.5 | 18th |
| Long | 62 | 61.2 | 74.6 | −13.4 | 16th |
By section of the resume
F1 by resume section, against the median system.
| Section | Resumes | F1 | Median | Gap |
|---|---|---|---|---|
| Contact details | 298 | 73.3 | 93.5 | −20.3 |
| Experience | 298 | 55.4 | 86.3 | −30.9 |
| Education | 296 | 64.2 | 89.4 | −25.2 |
| Summary | 288 | 50.7 | 81.7 | −31.0 |
| Certifications | 244 | 57.3 | 81.1 | −23.8 |
| Awards | 103 | 54.3 | 75.4 | −21.1 |
| Skills | 285 | 66.0 | 92.1 | −26.1 |
| Languages | 178 | 65.8 | 84.3 | −18.5 |
| Custom sections | 212 | 32.0 | 49.9 | −17.9 |
| Interests | 74 | 44.3 | 91.7 | −47.4 |
| Volunteering | 79 | 43.0 | 84.5 | −41.6 |
| Profiles | 38 | 41.2 | 66.7 | −25.5 |
| Projects | 81 | 60.5 | 83.2 | −22.7 |
| Publications | 50 | 65.0 | 78.6 | −13.6 |
On the published resumes
Best and worst of the 20 public resumes.
| Resume | What makes it hard | F1 |
|---|---|---|
| Creative Director | Accented text in a styled CV | 91.4 |
| Lead Respiratory Therapist (Adult ICU) | Long resume, 34 entries | 89.0 |
| Deputy Head of Mission | Rotated scan with hidden identity text | 87.0 |
| Head of Events & Public Space Operations | Table grid with emoji bullets | 10.8 |
| Partner | Consulting CV, six pages | 10.3 |
| Quality Systems Lead | Photo of a printout | 9.2 |
How it was run
Z.ai API, GLM-5.3-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 |
|---|---|---|---|---|
| Qwen 3.5 35B | 77.9 | $3.91 | 78 s | GLM 5.3 Flash vs Qwen 3.5 35B |
| Qwen 3.5 9B | 76.0 | $1.31 | 138 s | GLM 5.3 Flash vs Qwen 3.5 9B |
| Gemma 4 26B | 62.8 | $0.56 | 29 s | GLM 5.3 Flash vs Gemma 4 26B |
| GPT-5.4 Nano | 61.9 | $2.76 | 20 s | GLM 5.3 Flash vs GPT-5.4 Nano |
GLM 5.3 Flash against every other model
- GLM 5.3 Flash vs LlamaParse Extract Agentic Plus
- GLM 5.3 Flash vs LlamaParse Extract Agentic
- GLM 5.3 Flash vs LlamaParse Extract Cost-Effective
- GLM 5.3 Flash vs Reducto Deep Extract
- GLM 5.3 Flash vs Reducto Extract
- GLM 5.3 Flash vs Extend Max Context
- GLM 5.3 Flash vs Extend Extract
- GLM 5.3 Flash vs GPT-5.6
- GLM 5.3 Flash vs GPT-5.5
- GLM 5.3 Flash vs Claude Opus 4.8
- GLM 5.3 Flash vs Gemini 3.5 Flash
- GLM 5.3 Flash vs Qwen 3.8 Flash
- GLM 5.3 Flash vs Qwen 3.8 27B
Embed this result
Shows its current rank. Updates with each run.