Anthropic · Vision LLM
Claude Opus 4.8 on ResumeBench: 85.5 F1, rank 4 of 18
Claude Opus 4.8 ranks 4th of 18 on ResumeBench: 85.5 F1, $90 per 1,000 pages, 31 seconds per resume. It is 1.7 points behind LlamaParse Extract Agentic Plus. LlamaParse Extract Agentic Plus scores 87.2 for $54.
1st of 11 vision LLMs. Strongest on hidden text & prompt injections (+6.3 points), image only (+5.7 points) and scanned & rotated (+4.6 points). Weakest on content as images (−4.6 points), wording traps (−2.6 points) and distractor layouts (−2.5 points).
- Rank
- 4/18
- F1
- 85.5
- Precision
- 86.8
- Recall
- 87.5
- Perfect resumes
- 31/300
- Per 1,000 pages
- $90
- Median per resume
- 31 s
Averaged over 300 resumes. Dates (scored separately): 97.7. 90% finish within 52 s.
Where Claude Opus 4.8 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 | 54.7 | 51.5 | +3.3 | 3rd |
| Image only | 28 | 77.0 | 71.4 | +5.7 | 3rd |
| Sidebar or grid | 82 | 84.2 | 81.9 | +2.3 | 7th |
| Photo | 45 | 86.3 | 84.0 | +2.2 | 4th |
Distractors
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Rotated & warped pages | 21 | 87.4 | 82.9 | +4.4 | 8th |
| Hidden text & prompt injections | 14 | 90.8 | 84.5 | +6.3 | 3rd |
| Content as images | 8 | 83.6 | 88.2 | −4.6 | 14th |
| Wording traps | 16 | 81.9 | 84.5 | −2.6 | 12th |
| Typos kept on purpose | 37 | 85.2 | 83.7 | +1.6 | 8th |
Extraction challenges
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Many entries | 83 | 83.7 | 81.9 | +1.8 | 7th |
| Long bullet lists | 101 | 83.1 | 81.4 | +1.7 | 7th |
| Split across pages | 85 | 84.6 | 82.2 | +2.4 | 4th |
| Big skills list | 60 | 84.8 | 83.6 | +1.2 | 7th |
| Custom sections | 105 | 83.4 | 80.8 | +2.6 | 3rd |
| Combined headings | 37 | 85.1 | 82.7 | +2.5 | 1st |
| Promotions | 88 | 85.5 | 83.5 | +2.0 | 6th |
| Partial dates | 138 | 84.5 | 83.4 | +1.1 | 7th |
| Overlapping jobs | 69 | 83.5 | 81.7 | +1.7 | 6th |
| Accented text | 13 | 85.6 | 85.5 | +0.1 | 9th |
Noise
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Repeated header | 67 | 87.6 | 85.2 | +2.4 | 2nd |
| References | 55 | 87.6 | 85.2 | +2.5 | 6th |
| Busy page | 18 | 80.8 | 80.1 | +0.7 | 7th |
Source
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Everyday resumes | 212 | 86.8 | 85.1 | +1.6 | 5th |
| Expert-built (Canva / LaTeX) | 32 | 86.1 | 83.6 | +2.5 | 2nd |
| Resume-builder templates | 10 | 85.4 | 81.2 | +4.2 | 4th |
| Scanned & rotated | 19 | 87.5 | 82.9 | +4.6 | 7th |
| Distractor layouts | 18 | 82.8 | 85.3 | −2.5 | 12th |
Layout
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Single column | 195 | 85.6 | 84.0 | +1.6 | 2nd |
| Two-column | 23 | 89.7 | 86.8 | +2.8 | 3rd |
| Two-column sidebar | 54 | 81.9 | 80.0 | +1.9 | 7th |
| Three-column | 20 | 86.2 | 84.8 | +1.4 | 2nd |
| Table grid | 8 | 95.1 | 94.1 | +1.1 | 6th |
Length
| Kind of resume | # of resumes | F1 | Median | Gap | Rank |
|---|---|---|---|---|---|
| Short | 148 | 88.3 | 87.1 | +1.1 | 5th |
| Medium | 90 | 85.8 | 84.7 | +1.2 | 3rd |
| Long | 62 | 78.4 | 74.6 | +3.8 | 3rd |
By section of the resume
F1 by resume section, against the median system.
| Section | Resumes | F1 | Median | Gap |
|---|---|---|---|---|
| Contact details | 300 | 93.7 | 93.5 | +0.2 |
| Experience | 300 | 88.2 | 86.3 | +1.9 |
| Education | 297 | 90.2 | 89.4 | +0.8 |
| Summary | 290 | 89.7 | 81.7 | +7.9 |
| Certifications | 239 | 85.4 | 81.1 | +4.3 |
| Awards | 106 | 72.9 | 75.4 | −2.4 |
| Skills | 283 | 95.2 | 92.1 | +3.1 |
| Languages | 167 | 85.8 | 84.3 | +1.5 |
| Custom sections | 218 | 51.6 | 49.9 | +1.7 |
| Interests | 63 | 93.7 | 91.7 | +1.9 |
| Volunteering | 78 | 84.6 | 84.5 | +0.1 |
| Profiles | 35 | 61.4 | 66.7 | −5.3 |
| Projects | 78 | 82.1 | 83.2 | −1.1 |
| Publications | 52 | 81.8 | 78.6 | +3.2 |
On the published resumes
Best and worst of the 20 public resumes.
| Resume | What makes it hard | F1 |
|---|---|---|
| Senior Manufacturing Engineer | Third-person prose, no bullets | 93.0 |
| Head of Developer Experience (Internal title: Jedi Master of Analytics) | Unusual job titles | 88.8 |
| Quality Systems Lead | Photo of a printout | 87.3 |
| Senior Game Designer | Game designer's Canva resume | 64.6 |
| Senior ocularist and practice owner | Handwritten resume, six pages | 49.7 |
| senior Consultant - package Implementation | Handwritten resume | 49.4 |
How it was run
Anthropic API, Claude Opus 4.8, 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 | Claude Opus 4.8 vs Reducto Deep Extract |
| LlamaParse Extract Agentic | 85.5 | $23 | 55 s | Claude Opus 4.8 vs LlamaParse Extract Agentic |
| Gemini 3.5 Flash | 85.5 | $35 | 43 s | Claude Opus 4.8 vs Gemini 3.5 Flash |
| GPT-5.6 | 84.7 | $74 | 38 s | Claude Opus 4.8 vs GPT-5.6 |
Claude Opus 4.8 against every other model
- Claude Opus 4.8 vs LlamaParse Extract Agentic Plus
- Claude Opus 4.8 vs LlamaParse Extract Cost-Effective
- Claude Opus 4.8 vs Reducto Extract
- Claude Opus 4.8 vs Extend Max Context
- Claude Opus 4.8 vs Extend Extract
- Claude Opus 4.8 vs GPT-5.5
- Claude Opus 4.8 vs GPT-5.4 Nano
- Claude Opus 4.8 vs Gemma 4 26B
- Claude Opus 4.8 vs Qwen 3.8 Flash
- Claude Opus 4.8 vs Qwen 3.8 27B
- Claude Opus 4.8 vs Qwen 3.5 35B
- Claude Opus 4.8 vs Qwen 3.5 9B
- Claude Opus 4.8 vs GLM 5.3 Flash
Embed this result
Shows its current rank. Updates with each run.