Skip to content

Alibaba · Vision LLM

Qwen 3.5 35B on ResumeBench: 77.9 F1, rank 14 of 18

Qwen 3.5 35B ranks 14th of 18 on ResumeBench: 77.9 F1, $3.91 per 1,000 pages, 78 seconds per resume. It is 9.2 points behind LlamaParse Extract Agentic Plus. Qwen 3.8 Flash scores 84.1 for $2.98.

7th of 11 vision LLMs. Strongest on content as images (+2.1 points), handwritten (+1.3 points) and table grid (+0.4 points). Weakest on two-column (−9.4 points), three-column (−9.2 points) and two-column sidebar (−8.5 points).

Rank
14/18
F1
77.9
Precision
78.8
Recall
83.7
Perfect resumes
4/300
Per 1,000 pages
$3.91
Median per resume
78 s

Averaged over 300 resumes. Dates (scored separately): 88.5. 90% finish within 221 s.

Where Qwen 3.5 35B wins and loses

F1 by kind of resume, against the median system.

Reading the page

Qwen 3.5 35B by reading the page
Kind of resume# of resumesF1MedianGapRank
Handwritten952.751.5+1.37th
Image only2869.771.4−1.712th
Sidebar or grid8274.881.9−7.214th
Photo4580.084.0−4.113th

Distractors

Qwen 3.5 35B by distractors
Kind of resume# of resumesF1MedianGapRank
Rotated & warped pages2176.582.9−6.412th
Hidden text & prompt injections1479.284.5−5.312th
Content as images890.388.2+2.15th
Wording traps1684.884.5+0.39th
Typos kept on purpose3780.483.7−3.313th

Extraction challenges

Qwen 3.5 35B by extraction challenges
Kind of resume# of resumesF1MedianGapRank
Many entries8378.981.9−2.914th
Long bullet lists10176.381.4−5.114th
Split across pages8576.582.2−5.714th
Big skills list6076.683.6−7.014th
Custom sections10577.080.8−3.814th
Combined headings3774.682.7−8.114th
Promotions8876.783.5−6.814th
Partial dates13878.983.4−4.614th
Overlapping jobs6976.281.7−5.514th
Accented text1380.485.5−5.115th

Noise

Qwen 3.5 35B by noise
Kind of resume# of resumesF1MedianGapRank
Repeated header6778.685.2−6.614th
References5581.885.2−3.414th
Busy page1871.880.1−8.314th

Source

Qwen 3.5 35B by source
Kind of resume# of resumesF1MedianGapRank
Everyday resumes21277.785.1−7.514th
Expert-built (Canva / LaTeX)3283.783.6+0.09th
Resume-builder templates1079.181.2−2.113th
Scanned & rotated1975.382.9−7.612th
Distractor layouts1885.785.3+0.47th

Layout

Qwen 3.5 35B by layout
Kind of resume# of resumesF1MedianGapRank
Single column19579.484.0−4.614th
Two-column2377.486.8−9.415th
Two-column sidebar5471.580.0−8.514th
Three-column2075.684.8−9.215th
Table grid894.594.1+0.49th

Length

Qwen 3.5 35B by length
Kind of resume# of resumesF1MedianGapRank
Short14879.887.1−7.315th
Medium9079.784.7−4.914th
Long6271.074.6−3.714th

By section of the resume

F1 by resume section, against the median system.

Qwen 3.5 35B by resume section
SectionResumesF1MedianGap
Contact details30092.193.5−1.4
Experience30079.386.3−7.1
Education29781.889.4−7.6
Summary29180.481.7−1.3
Certifications25179.181.1−2.0
Awards11661.975.4−13.5
Skills29487.892.1−4.3
Languages17579.484.3−4.9
Custom sections23640.249.9−9.7
Interests6689.791.7−2.0
Volunteering8083.184.5−1.4
Profiles3965.866.7−0.9
Projects8176.783.2−6.4
Publications5375.078.6−3.6

On the published resumes

Best and worst of the 20 public resumes.

Qwen 3.5 35B on published resumes
ResumeWhat makes it hardF1
Head of Developer Experience (Internal title: Jedi Master of Analytics) Unusual job titles90.6
Quality Systems LeadPhoto of a printout88.9
Head of Events & Public Space OperationsTable grid with emoji bullets88.2
Senior Game DesignerGame designer's Canva resume66.8
senior Consultant - package ImplementationHandwritten resume59.2
Senior ocularist and practice ownerHandwritten resume, six pages38.5

How it was run

Qwen3.5-35B-A3B (open weights), page images in, JSON schema out. Same PDF, schema and instructions as every other system, one run per resume (4 November 2026).

Qwen 3.5 35B on Alibaba’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.

Qwen 3.5 35B: #14 on ResumeBench resume extraction benchmark
<a href="https://www.resumebench.ai/systems/qwen-3-5-35b"><img src="https://www.resumebench.ai/badge/qwen-3-5-35b.svg" alt="Qwen 3.5 35B: #14 on ResumeBench resume extraction benchmark" height="28"></a>

Choosing a resume extraction model

All 18 models · Leaderboard

Frequently asked questions (FAQs)

How accurate is Qwen 3.5 35B at resume extraction?

77.9 F1 on ResumeBench's 300 resumes, 14th of 18 models.

How much does Qwen 3.5 35B cost to parse resumes?

$3.91 per 1,000 pages, about $0.011 per resume.

How fast is Qwen 3.5 35B?

78 seconds per resume (median).

Which resumes is Qwen 3.5 35B best at?

Content as images, handwritten and table grid. Weakest on two-column, three-column and two-column sidebar.

What is a cheaper alternative to Qwen 3.5 35B?

Qwen 3.8 Flash: 84.1 F1 for $2.98 per 1,000 pages.