Google · Vision LLM
Gemini 3.5 Flash on ResumeBench: 85.5 F1, rank 5 of 18
Gemini 3.5 Flash ranks 5th of 18 on ResumeBench: 85.5 F1, $35 per 1,000 pages, 43 seconds per resume. It is 1.7 points behind LlamaParse Extract Agentic Plus. LlamaParse Extract Agentic scores 85.5 for $23.
2nd of 11 vision LLMs. Strongest on references (+5.5 points), resume-builder templates (+5.2 points) and overlapping jobs (+4.7 points). Weakest on content as images (−11.5 points), wording traps (−4.1 points) and distractor layouts (−4.0 points).
- Rank
- 5/18
- F1
- 85.5
- Precision
- 86.9
- Recall
- 87.9
- Perfect resumes
- 34/300
- Per 1,000 pages
- $35
- Median per resume
- 43 s
Averaged over 300 resumes. Dates (scored separately): 97.8. 90% finish within 70 s.
Where Gemini 3.5 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 | 50.3 | 51.5 | −1.2 | 13th |
| Image only | 28 | 75.1 | 71.4 | +3.7 | 7th |
| Sidebar or grid | 82 | 85.1 | 81.9 | +3.1 | 4th |
| Photo | 45 | 85.1 | 84.0 | +1.0 | 7th |
Distractors
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Rotated & warped pages | 21 | 87.4 | 82.9 | +4.5 | 7th |
| Hidden text & prompt injections | 14 | 86.7 | 84.5 | +2.1 | 9th |
| Content as images | 8 | 76.7 | 88.2 | −11.5 | 15th |
| Wording traps | 16 | 80.4 | 84.5 | −4.1 | 14th |
| Typos kept on purpose | 37 | 84.4 | 83.7 | +0.7 | 9th |
Extraction challenges
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Many entries | 83 | 84.3 | 81.9 | +2.4 | 5th |
| Long bullet lists | 101 | 83.9 | 81.4 | +2.4 | 6th |
| Split across pages | 85 | 84.6 | 82.2 | +2.4 | 5th |
| Big skills list | 60 | 86.0 | 83.6 | +2.4 | 1st |
| Custom sections | 105 | 83.6 | 80.8 | +2.7 | 2nd |
| Combined headings | 37 | 84.2 | 82.7 | +1.5 | 3rd |
| Promotions | 88 | 87.3 | 83.5 | +3.9 | 2nd |
| Partial dates | 138 | 84.9 | 83.4 | +1.5 | 5th |
| Overlapping jobs | 69 | 86.4 | 81.7 | +4.7 | 1st |
| Accented text | 13 | 86.4 | 85.5 | +0.8 | 8th |
Noise
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Repeated header | 67 | 86.3 | 85.2 | +1.2 | 6th |
| References | 55 | 90.7 | 85.2 | +5.5 | 1st |
| Busy page | 18 | 80.8 | 80.1 | +0.7 | 8th |
Source
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Everyday resumes | 212 | 87.2 | 85.1 | +2.1 | 2nd |
| Expert-built (Canva / LaTeX) | 32 | 84.6 | 83.6 | +1.0 | 5th |
| Resume-builder templates | 10 | 86.4 | 81.2 | +5.2 | 2nd |
| Scanned & rotated | 19 | 87.2 | 82.9 | +4.3 | 8th |
| Distractor layouts | 18 | 81.4 | 85.3 | −4.0 | 14th |
Layout
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Single column | 195 | 85.1 | 84.0 | +1.1 | 6th |
| Two-column | 23 | 90.0 | 86.8 | +3.2 | 2nd |
| Two-column sidebar | 54 | 83.4 | 80.0 | +3.4 | 4th |
| Three-column | 20 | 85.6 | 84.8 | +0.8 | 5th |
| Table grid | 8 | 95.0 | 94.1 | +0.9 | 7th |
Length
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Short | 148 | 88.3 | 87.1 | +1.2 | 4th |
| Medium | 90 | 85.0 | 84.7 | +0.4 | 8th |
| Long | 62 | 79.2 | 74.6 | +4.5 | 2nd |
By section of the resume
F1 by resume section, against the median system.
| Section | Resumes | F1 | Median | Gap |
|---|---|---|---|---|
| Contact details | 300 | 95.3 | 93.5 | +1.7 |
| Experience | 300 | 86.5 | 86.3 | +0.2 |
| Education | 297 | 91.5 | 89.4 | +2.1 |
| Summary | 290 | 89.3 | 81.7 | +7.6 |
| Certifications | 245 | 84.0 | 81.1 | +2.8 |
| Awards | 92 | 80.3 | 75.4 | +4.9 |
| Skills | 289 | 93.5 | 92.1 | +1.4 |
| Languages | 170 | 84.3 | 84.3 | −0.0 |
| Custom sections | 216 | 50.1 | 49.9 | +0.2 |
| Interests | 63 | 94.2 | 91.7 | +2.4 |
| Volunteering | 79 | 83.7 | 84.5 | −0.8 |
| Profiles | 35 | 64.8 | 66.7 | −2.0 |
| Projects | 80 | 80.4 | 83.2 | −2.8 |
| Publications | 52 | 83.4 | 78.6 | +4.8 |
On the published resumes
Best and worst of the 20 public resumes.
| Resume | What makes it hard | F1 |
|---|---|---|
| Lead Respiratory Therapist (Adult ICU) | Long resume, 34 entries | 96.7 |
| Senior Data Scientist (Lead) | Fifty skills, seven pages | 95.8 |
| Senior Manufacturing Engineer | Third-person prose, no bullets | 93.4 |
| Senior Game Designer | Game designer's Canva resume | 54.8 |
| senior Consultant - package Implementation | Handwritten resume | 42.1 |
| Senior ocularist and practice owner | Handwritten resume, six pages | 39.8 |
How it was run
Gemini API, Gemini 3.5 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 |
|---|---|---|---|---|
| Reducto Deep Extract | 85.6 | $40 | 65 s | Gemini 3.5 Flash vs Reducto Deep Extract |
| LlamaParse Extract Agentic | 85.5 | $23 | 55 s | Gemini 3.5 Flash vs LlamaParse Extract Agentic |
| Claude Opus 4.8 | 85.5 | $90 | 31 s | Gemini 3.5 Flash vs Claude Opus 4.8 |
| GPT-5.6 | 84.7 | $74 | 38 s | Gemini 3.5 Flash vs GPT-5.6 |
Gemini 3.5 Flash against every other model
- Gemini 3.5 Flash vs LlamaParse Extract Agentic Plus
- Gemini 3.5 Flash vs LlamaParse Extract Cost-Effective
- Gemini 3.5 Flash vs Reducto Extract
- Gemini 3.5 Flash vs Extend Max Context
- Gemini 3.5 Flash vs Extend Extract
- Gemini 3.5 Flash vs GPT-5.5
- Gemini 3.5 Flash vs GPT-5.4 Nano
- Gemini 3.5 Flash vs Gemma 4 26B
- Gemini 3.5 Flash vs Qwen 3.8 Flash
- Gemini 3.5 Flash vs Qwen 3.8 27B
- Gemini 3.5 Flash vs Qwen 3.5 35B
- Gemini 3.5 Flash vs Qwen 3.5 9B
- Gemini 3.5 Flash vs GLM 5.3 Flash
Embed this result
Shows its current rank. Updates with each run.