Skip to content

OpenAI · Vision LLM

GPT-5.5 on ResumeBench: 83.5 F1, rank 11 of 18

GPT-5.5 ranks 11th of 18 on ResumeBench: 83.5 F1, $87 per 1,000 pages, 48 seconds per resume. It is 3.7 points behind LlamaParse Extract Agentic Plus. LlamaParse Extract Agentic Plus scores 87.2 for $54.

6th of 11 vision LLMs. Strongest on content as images (+4.3 points), references (+3.3 points) and long bullet lists (+2.9 points). Weakest on scanned & rotated (−22.2 points), rotated & warped pages (−19.6 points) and hidden text & prompt injections (−15.4 points).

Rank
11/18
F1
83.5
Precision
84.9
Recall
86.0
Perfect resumes
32/300
Per 1,000 pages
$87
Median per resume
48 s

Averaged over 300 resumes. Dates (scored separately): 97.8. 90% finish within 75 s.

Where GPT-5.5 wins and loses

F1 by kind of resume, against the median system.

Reading the page

GPT-5.5 by reading the page
Kind of resume# of resumesF1MedianGapRank
Handwritten951.051.5−0.511th
Image only2857.071.4−14.415th
Sidebar or grid8280.481.9−1.512th
Photo4584.084.0−0.010th

Distractors

GPT-5.5 by distractors
Kind of resume# of resumesF1MedianGapRank
Rotated & warped pages2163.482.9−19.615th
Hidden text & prompt injections1469.184.5−15.415th
Content as images892.688.2+4.33rd
Wording traps1685.384.5+0.87th
Typos kept on purpose3773.383.7−10.414th

Extraction challenges

GPT-5.5 by extraction challenges
Kind of resume# of resumesF1MedianGapRank
Many entries8383.881.9+1.96th
Long bullet lists10184.381.4+2.95th
Split across pages8585.082.2+2.82nd
Big skills list6085.583.6+1.95th
Custom sections10580.880.8−0.010th
Combined headings3783.482.7+0.76th
Promotions8884.083.5+0.58th
Partial dates13884.083.4+0.68th
Overlapping jobs6982.181.7+0.39th
Accented text1383.985.5−1.713th

Noise

GPT-5.5 by noise
Kind of resume# of resumesF1MedianGapRank
Repeated header6786.785.2+1.55th
References5588.585.2+3.34th
Busy page1882.580.1+2.46th

Source

GPT-5.5 by source
Kind of resume# of resumesF1MedianGapRank
Everyday resumes21286.685.1+1.56th
Expert-built (Canva / LaTeX)3283.683.6−0.010th
Resume-builder templates1083.981.2+2.76th
Scanned & rotated1960.782.9−22.215th
Distractor layouts1886.685.3+1.26th

Layout

GPT-5.5 by layout
Kind of resume# of resumesF1MedianGapRank
Single column19584.284.0+0.28th
Two-column2388.386.8+1.56th
Two-column sidebar5477.280.0−2.813th
Three-column2082.684.8−2.211th
Table grid896.894.1+2.72nd

Length

GPT-5.5 by length
Kind of resume# of resumesF1MedianGapRank
Short14885.287.1−1.912th
Medium9084.984.7+0.29th
Long6277.374.6+2.77th

By section of the resume

F1 by resume section, against the median system.

GPT-5.5 by resume section
SectionResumesF1MedianGap
Contact details30093.593.5−0.1
Experience30087.486.3+1.1
Education29788.789.4−0.7
Summary29181.881.7+0.1
Certifications25477.381.1−3.8
Awards9875.375.4−0.1
Skills28791.592.1−0.6
Languages16885.784.3+1.4
Custom sections21049.749.9−0.2
Interests6492.491.7+0.7
Volunteering7785.084.5+0.5
Profiles3668.566.7+1.8
Projects7986.483.2+3.2
Publications5182.578.6+3.9

On the published resumes

Best and worst of the 20 public resumes.

GPT-5.5 on published resumes
ResumeWhat makes it hardF1
Head of Events & Public Space OperationsTable grid with emoji bullets99.5
Creative DirectorAccented text in a styled CV90.2
Lead Respiratory Therapist (Adult ICU)Long resume, 34 entries88.7
Deputy Head of MissionRotated scan with hidden identity text64.7
senior Consultant - package ImplementationHandwritten resume51.7
Senior ocularist and practice ownerHandwritten resume, six pages37.5

How it was run

OpenAI API, model gpt-5.5, page images in, JSON schema out. Same PDF, schema and instructions as every other system, one run per resume (4 November 2026).

GPT-5.5 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-5.5: #11 on ResumeBench resume extraction benchmark
<a href="https://www.resumebench.ai/systems/gpt-5-5"><img src="https://www.resumebench.ai/badge/gpt-5-5.svg" alt="GPT-5.5: #11 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-5.5 at resume extraction?

83.5 F1 on ResumeBench's 300 resumes, 11th of 18 models.

How much does GPT-5.5 cost to parse resumes?

$87 per 1,000 pages, about $0.244 per resume.

How fast is GPT-5.5?

48 seconds per resume (median).

Which resumes is GPT-5.5 best at?

Content as images, references and long bullet lists. Weakest on scanned & rotated, rotated & warped pages and hidden text & prompt injections.

What is a cheaper alternative to GPT-5.5?

LlamaParse Extract Agentic Plus: 87.2 F1 for $54 per 1,000 pages.