Skip to content

Alibaba · Vision LLM

Qwen 3.5 9B on ResumeBench: 76.0 F1, rank 15 of 18

Qwen 3.5 9B ranks 15th of 18 on ResumeBench: 76.0 F1, $1.31 per 1,000 pages, 138 seconds per resume. It is 11.1 points behind LlamaParse Extract Agentic Plus.

8th of 11 vision LLMs. Strongest on table grid (−0.8 points), handwritten (−0.9 points) and accented text (−1.8 points). Weakest on scanned & rotated (−18.4 points), rotated & warped pages (−16.5 points) and hidden text & prompt injections (−14.4 points).

Rank
15/18
F1
76.0
Precision
78.9
Recall
80.3
Perfect resumes
6/300
Per 1,000 pages
$1.31
Median per resume
138 s

Averaged over 300 resumes. Dates (scored separately): 70.4. 90% finish within 490 s.

Where Qwen 3.5 9B wins and loses

F1 by kind of resume, against the median system.

Reading the page

Qwen 3.5 9B by reading the page
Kind of resume# of resumesF1MedianGapRank
Handwritten950.651.5−0.912th
Image only2861.271.4−10.114th
Sidebar or grid8273.581.9−8.415th
Photo4574.984.0−9.215th

Distractors

Qwen 3.5 9B by distractors
Kind of resume# of resumesF1MedianGapRank
Rotated & warped pages2166.582.9−16.514th
Hidden text & prompt injections1470.284.5−14.414th
Content as images885.988.2−2.312th
Wording traps1677.784.5−6.815th
Typos kept on purpose3772.283.7−11.415th

Extraction challenges

Qwen 3.5 9B by extraction challenges
Kind of resume# of resumesF1MedianGapRank
Many entries8373.981.9−8.015th
Long bullet lists10173.281.4−8.215th
Split across pages8573.482.2−8.815th
Big skills list6076.383.6−7.315th
Custom sections10573.080.8−7.815th
Combined headings3772.282.7−10.515th
Promotions8874.883.5−8.715th
Partial dates13875.683.4−7.915th
Overlapping jobs6972.281.7−9.615th
Accented text1383.785.5−1.814th

Noise

Qwen 3.5 9B by noise
Kind of resume# of resumesF1MedianGapRank
Repeated header6776.485.2−8.715th
References5577.685.2−7.615th
Busy page1867.680.1−12.515th

Source

Qwen 3.5 9B by source
Kind of resume# of resumesF1MedianGapRank
Everyday resumes21277.285.1−7.915th
Expert-built (Canva / LaTeX)3279.183.6−4.514th
Resume-builder templates1078.581.2−2.714th
Scanned & rotated1964.582.9−18.414th
Distractor layouts1880.485.3−5.015th

Layout

Qwen 3.5 9B by layout
Kind of resume# of resumesF1MedianGapRank
Single column19576.784.0−7.215th
Two-column2379.186.8−7.714th
Two-column sidebar5469.780.0−10.315th
Three-column2075.884.8−9.014th
Table grid893.394.1−0.811th

Length

Qwen 3.5 9B by length
Kind of resume# of resumesF1MedianGapRank
Short14878.187.1−9.016th
Medium9078.084.7−6.715th
Long6268.074.6−6.615th

By section of the resume

F1 by resume section, against the median system.

Qwen 3.5 9B by resume section
SectionResumesF1MedianGap
Contact details29992.693.5−1.0
Experience29977.286.3−9.1
Education29680.889.4−8.7
Summary29077.981.7−3.8
Certifications24674.781.1−6.4
Awards12755.275.4−20.2
Skills29383.292.1−8.9
Languages17280.084.3−4.3
Custom sections21436.449.9−13.5
Interests6884.391.7−7.4
Volunteering8080.284.5−4.3
Profiles3851.366.7−15.4
Projects7975.583.2−7.7
Publications5376.378.6−2.2

On the published resumes

Best and worst of the 20 public resumes.

Qwen 3.5 9B on published resumes
ResumeWhat makes it hardF1
Head of Events & Public Space OperationsTable grid with emoji bullets88.2
Head of Digital MarketingCanva design with a photo85.2
Quality Systems LeadPhoto of a printout79.4
Supervisory Human Resources Specialist, GS-0201-15Resume-builder template54.9
senior Consultant - package ImplementationHandwritten resume48.7
Senior ocularist and practice ownerHandwritten resume, six pages43.7

How it was run

Qwen3.5-9B (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 9B 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 9B: #15 on ResumeBench resume extraction benchmark
<a href="https://www.resumebench.ai/systems/qwen-3-5-9b"><img src="https://www.resumebench.ai/badge/qwen-3-5-9b.svg" alt="Qwen 3.5 9B: #15 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 9B at resume extraction?

76.0 F1 on ResumeBench's 300 resumes, 15th of 18 models.

How much does Qwen 3.5 9B cost to parse resumes?

$1.31 per 1,000 pages, about $0.004 per resume.

How fast is Qwen 3.5 9B?

138 seconds per resume (median).

Which resumes is Qwen 3.5 9B best at?

Table grid, handwritten and accented text. Weakest on scanned & rotated, rotated & warped pages and hidden text & prompt injections.

What is a cheaper alternative to Qwen 3.5 9B?

Gemma 4 26B: 62.8 F1 for $0.56 per 1,000 pages.