Skip to content

Alibaba · Vision LLM

Qwen 3.8 Flash on ResumeBench: 84.1 F1, rank 8 of 18

Qwen 3.8 Flash ranks 8th of 18 on ResumeBench: 84.1 F1, $2.98 per 1,000 pages, 122 seconds per resume. It is 3.1 points behind LlamaParse Extract Agentic Plus.

4th of 11 vision LLMs. Strongest on hidden text & prompt injections (+4.5 points), scanned & rotated (+3.4 points) and image only (+3.3 points). Weakest on overlapping jobs (−0.3 points), busy page (−0.3 points) and distractor layouts (−0.2 points).

Rank
8/18
F1
84.1
Precision
85.4
Recall
86.2
Perfect resumes
25/300
Per 1,000 pages
$2.98
Median per resume
122 s

Averaged over 300 resumes. Dates (scored separately): 76.5. 90% finish within 246 s.

Where Qwen 3.8 Flash wins and loses

F1 by kind of resume, against the median system.

Reading the page

Qwen 3.8 Flash by reading the page
Kind of resume# of resumesF1MedianGapRank
Handwritten953.551.5+2.05th
Image only2874.771.4+3.38th
Sidebar or grid8283.881.9+1.98th
Photo4586.384.0+2.33rd

Distractors

Qwen 3.8 Flash by distractors
Kind of resume# of resumesF1MedianGapRank
Rotated & warped pages2186.082.9+3.19th
Hidden text & prompt injections1489.084.5+4.57th
Content as images888.388.2+0.08th
Wording traps1686.084.5+1.55th
Typos kept on purpose3785.783.7+2.17th

Extraction challenges

Qwen 3.8 Flash by extraction challenges
Kind of resume# of resumesF1MedianGapRank
Many entries8382.181.9+0.29th
Long bullet lists10182.181.4+0.69th
Split across pages8582.682.2+0.49th
Big skills list6083.583.6−0.110th
Custom sections10580.880.8+0.09th
Combined headings3782.682.7−0.010th
Promotions8883.883.5+0.39th
Partial dates13883.383.4−0.110th
Overlapping jobs6981.481.7−0.310th
Accented text1386.985.5+1.46th

Noise

Qwen 3.8 Flash by noise
Kind of resume# of resumesF1MedianGapRank
Repeated header6785.185.2−0.110th
References5585.485.2+0.29th
Busy page1879.880.1−0.310th

Source

Qwen 3.8 Flash by source
Kind of resume# of resumesF1MedianGapRank
Everyday resumes21285.185.1−0.110th
Expert-built (Canva / LaTeX)3284.483.6+0.77th
Resume-builder templates1084.081.2+2.95th
Scanned & rotated1986.382.9+3.49th
Distractor layouts1885.185.3−0.211th

Layout

Qwen 3.8 Flash by layout
Kind of resume# of resumesF1MedianGapRank
Single column19583.884.0−0.110th
Two-column2387.186.8+0.38th
Two-column sidebar5481.680.0+1.69th
Three-column2084.984.8+0.19th
Table grid895.594.1+1.44th

Length

Qwen 3.8 Flash by length
Kind of resume# of resumesF1MedianGapRank
Short14887.287.1+0.18th
Medium9085.384.7+0.66th
Long6275.074.6+0.39th

By section of the resume

F1 by resume section, against the median system.

Qwen 3.8 Flash by resume section
SectionResumesF1MedianGap
Contact details29893.693.5+0.1
Experience29888.086.3+1.7
Education29590.189.4+0.7
Summary28881.381.7−0.5
Certifications24882.281.1+1.1
Awards9475.575.4+0.1
Skills28590.092.1−2.1
Languages16784.684.3+0.3
Custom sections21053.949.9+4.0
Interests6391.191.7−0.6
Volunteering7883.984.5−0.6
Profiles3573.366.7+6.6
Projects7983.783.2+0.5
Publications5181.878.6+3.3

On the published resumes

Best and worst of the 20 public resumes.

Qwen 3.8 Flash on published resumes
ResumeWhat makes it hardF1
Head of Events & Public Space OperationsTable grid with emoji bullets96.3
Senior Customer Service RepresentativeBusy Canva page94.9
Head of Developer Experience (Internal title: Jedi Master of Analytics) Unusual job titles88.8
Senior Game DesignerGame designer's Canva resume67.7
senior Consultant - package ImplementationHandwritten resume44.9
Senior ocularist and practice ownerHandwritten resume, six pages40.2

How it was run

Alibaba Cloud Model Studio, qwen3.8-flash, page images in, JSON schema out. Same PDF, schema and instructions as every other system, one run per resume (4 November 2026).

Qwen 3.8 Flash 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 3.8 Flash: #8 on ResumeBench resume extraction benchmark
<a href="https://www.resumebench.ai/systems/qwen-3-8-flash"><img src="https://www.resumebench.ai/badge/qwen-3-8-flash.svg" alt="Qwen 3.8 Flash: #8 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 3.8 Flash at resume extraction?

84.1 F1 on ResumeBench's 300 resumes, 8th of 18 models.

How much does Qwen 3.8 Flash cost to parse resumes?

$2.98 per 1,000 pages, about $0.008 per resume.

How fast is Qwen 3.8 Flash?

122 seconds per resume (median).

Which resumes is Qwen 3.8 Flash best at?

Hidden text & prompt injections, scanned & rotated and image only. Weakest on overlapping jobs, busy page and distractor layouts.

What is a cheaper alternative to Qwen 3.8 Flash?

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