Skip to content

OpenAI · Vision LLM

GPT-5.4 Nano on ResumeBench: 61.9 F1, rank 17 of 18

GPT-5.4 Nano ranks 17th of 18 on ResumeBench: 61.9 F1, $2.76 per 1,000 pages, 20 seconds per resume. It is 25.3 points behind LlamaParse Extract Agentic Plus. Qwen 3.5 9B scores 76.0 for $1.31.

10th of 11 vision LLMs. Strongest on handwritten (−0.1 points), resume-builder templates (−13.6 points) and long (−15.6 points). Weakest on scanned & rotated (−76.6 points), rotated & warped pages (−71.3 points) and hidden text & prompt injections (−62.6 points).

Rank
17/18
F1
61.9
Precision
70.1
Recall
66.3
Perfect resumes
1/300
Per 1,000 pages
$2.76
Median per resume
20 s

Averaged over 300 resumes. Dates (scored separately): 86.7. 90% finish within 34 s.

Where GPT-5.4 Nano wins and loses

F1 by kind of resume, against the median system.

Reading the page

GPT-5.4 Nano by reading the page
Kind of resume# of resumesF1MedianGapRank
Handwritten951.351.5−0.110th
Image only2821.071.4−50.418th
Sidebar or grid8254.981.9−27.118th
Photo4559.484.0−24.718th

Distractors

GPT-5.4 Nano by distractors
Kind of resume# of resumesF1MedianGapRank
Rotated & warped pages2111.782.9−71.318th
Hidden text & prompt injections1422.084.5−62.618th
Content as images861.388.2−26.917th
Wording traps1662.884.5−21.716th
Typos kept on purpose3733.583.7−50.118th

Extraction challenges

GPT-5.4 Nano by extraction challenges
Kind of resume# of resumesF1MedianGapRank
Many entries8359.781.9−22.216th
Long bullet lists10161.081.4−20.416th
Split across pages8561.782.2−20.516th
Big skills list6064.983.6−18.716th
Custom sections10559.580.8−21.316th
Combined headings3764.682.7−18.117th
Promotions8858.783.5−24.818th
Partial dates13862.683.4−20.917th
Overlapping jobs6958.681.7−23.116th
Accented text1355.985.5−29.617th

Noise

GPT-5.4 Nano by noise
Kind of resume# of resumesF1MedianGapRank
Repeated header6764.285.2−21.017th
References5567.185.2−18.116th
Busy page1861.680.1−18.516th

Source

GPT-5.4 Nano by source
Kind of resume# of resumesF1MedianGapRank
Everyday resumes21267.285.1−17.916th
Expert-built (Canva / LaTeX)3260.383.6−23.317th
Resume-builder templates1067.681.2−13.616th
Scanned & rotated196.382.9−76.618th
Distractor layouts1862.285.3−23.117th

Layout

GPT-5.4 Nano by layout
Kind of resume# of resumesF1MedianGapRank
Single column19563.884.0−20.216th
Two-column2370.286.8−16.616th
Two-column sidebar5447.480.0−32.618th
Three-column2067.284.8−17.616th
Table grid874.494.1−19.718th

Length

GPT-5.4 Nano by length
Kind of resume# of resumesF1MedianGapRank
Short14864.187.1−23.118th
Medium9060.284.7−24.517th
Long6259.074.6−15.617th

By section of the resume

F1 by resume section, against the median system.

GPT-5.4 Nano by resume section
SectionResumesF1MedianGap
Contact details30081.293.5−12.4
Experience30063.786.3−22.6
Education29762.289.4−27.2
Summary29071.781.7−10.0
Certifications24453.681.1−27.5
Awards10244.975.4−30.5
Skills29174.892.1−17.3
Languages17069.084.3−15.3
Custom sections23126.349.9−23.6
Interests6670.091.7−21.7
Volunteering8059.984.5−24.7
Profiles6623.766.7−43.0
Projects8148.883.2−34.4
Publications5130.778.6−47.9

On the published resumes

Best and worst of the 20 public resumes.

GPT-5.4 Nano on published resumes
ResumeWhat makes it hardF1
Senior Customer Service RepresentativeBusy Canva page81.0
Lead Solutions ArchitectIndian format with CGPA77.6
Senior Manufacturing EngineerThird-person prose, no bullets75.2
Supervisory Human Resources Specialist, GS-0201-15Resume-builder template32.5
PartnerConsulting CV, six pages27.8
Deputy Head of MissionRotated scan with hidden identity text1.7

How it was run

OpenAI API, model gpt-5.4-nano, page images in, JSON schema out. Same PDF, schema and instructions as every other system, one run per resume (4 November 2026).

GPT-5.4 Nano on OpenAI’s site

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 test

Closest alternatives

Nearest in F1.

Embed this result

Shows its current rank. Updates with each run.

GPT-5.4 Nano: #17 on ResumeBench resume extraction benchmark
<a href="https://www.resumebench.ai/systems/gpt-5-4-nano"><img src="https://www.resumebench.ai/badge/gpt-5-4-nano.svg" alt="GPT-5.4 Nano: #17 on ResumeBench resume extraction benchmark" height="28"></a>

Choosing a resume extraction model

All 18 models · Leaderboard

Frequently asked questions (FAQs)

How accurate is GPT-5.4 Nano at resume extraction?

61.9 F1 on ResumeBench's 300 resumes, 17th of 18 models.

How much does GPT-5.4 Nano cost to parse resumes?

$2.76 per 1,000 pages, about $0.008 per resume.

How fast is GPT-5.4 Nano?

20 seconds per resume (median).

Which resumes is GPT-5.4 Nano best at?

Handwritten, resume-builder templates and long. Weakest on scanned & rotated, rotated & warped pages and hidden text & prompt injections.

What is a cheaper alternative to GPT-5.4 Nano?

Qwen 3.5 9B: 76.0 F1 for $1.31 per 1,000 pages.