DEV Community: li wujie

I Tested GLM-5.3-Flash and Qwen3.8-Flash on 24 Real Tasks

li wujie — Thu, 27 Aug 2026 14:09:47 +0000

I test-ran both of this week's open-weight flash models against 24 small, real workloads from an actual product stack — structured extraction, SEO metadata, and code fixes — and graded everything programmatically. The short version: on quality the two models are effectively tied, per-task cost lands within ~3%, and the biggest practical difference wasn't intelligence at all — it was a 429 storm.

Both models dropped within a day of each other. GLM-5.3-Flash (320B total, 18B active) rode a wave of "frontier intelligence at flash cost" claims, including a now-confirmed anonymous stint as ox-alpha on OpenRouter. Qwen3.8-Flash-Next (125B + 51B N-gram embedding, 6B active) is the architecture preview for Qwen4. On paper they're close: 63.4 vs 58.7 on DeepSWE 1.1. Benchmarks won't settle which one you should actually wire into a product this month. Small, boring, real tasks might.

The setup

Both via OpenRouter, temperature: 0, one call per task, same prompts: z-ai/glm-5.3-flash ($0.075/M in, $0.25/M out) and qwen/qwen3.8-flash ($0.15/M in, $0.47/M out).
3 suites, 24 tasks total, all graded by machine, not vibes:
- Extraction (10): messy real-world blobs — support tickets, log fragments, changelogs, pricing pages, review text — into strict JSON schemas.
- SEO metadata (10): title + meta description under hard constraints (title ≤ 60 chars, description 140–160 chars, keyword must appear in both).
- Code (4): write or fix small utility functions (slugify with accent transliteration, a buggy debounce, duration parsing, CSV escaping). Pass/fail = does the hidden test suite run green.
Everything recorded: latency, tokens, cost per call, retry counts.

One honesty note before the numbers: this is one day, one shared API pool, N=24 — not a benchmark lab. Treat it as a smoke test from a builder's desk, which is exactly the decision most people are making this week.

Finding 1: Extraction is a tie — and the "failures" are judgment calls, not errors

Both models returned schema-valid JSON on 10/10 tasks. Zero parse failures either way.

Under strict field-by-field grading against my gold answers, it looks lopsided — until you read the "failures":

Task	What actually happened
Phone-mount listing	Both extracted the product fine. GLM appended a `black` tag; Qwen kept `(BLACK)` in the name. Both defensible readings of the same sloppy source.
Support-ticket triage	Qwen escalated severity to `critical` where I expected `high`. With "URGENT!!!" and lost revenue in the ticket, that's a judgment call, not an error.
Changelog parse	My gold answer was the wrong one — both models correctly captured a BREAKING API removal I'd missed when writing the key. Both models beat my grader.

Regraded with a two-tier rubric (exact on objective fields; "defensible variant" for judgment fields like verdicts and free-text): GLM 4 exact + 4 defensible + 2 real misses; Qwen 3 exact + 6 defensible + 1 real miss. That's inside the noise floor.

Takeaway for builders: on these two models, extraction is a solved problem at the schema level. Your prompt design — how you pin enums, nulls, and array semantics — matters more than which of the two you pick.

Finding 2: Hard constraints — also a tie, but read the token meters

Both models hit 10/10 on the SEO-metadata suite: keyword present, title ≤ 60 chars, description inside the 140–160 window, every task.

The interesting part is what it cost to get there. Measured completion tokens on the same suite:

GLM-5.3-Flash: ~3,100 output tokens per task
Qwen3.8-Flash: ~1,500 output tokens per task

GLM burns roughly 2× the tokens to land the same two-line answer — presumably reasoning overhead — while charging half the per-token price. The two effects cancel almost exactly: $0.00048 per task for GLM vs $0.00049 for Qwen, within ~3% of each other across the full run. Price-per-token is half the story; tokens-per-task is the other half, and only their product shows up on your invoice.

Finding 3: Code — GLM 4/4, Qwen never got to answer

GLM-5.3-Flash passed all four code tasks first-try: the transliteration-aware slugify, the debounce fix (early-fire bug), duration parsing with error handling, and CSV escaping with quote-doubling.

Qwen3.8-Flash didn't fail these tasks. It never got to attempt them. See below.

Finding 4: The difference nobody benchmarks — availability

Here's what dominated the actual experience of running this test:

GLM-5.3-Flash: 24 calls, 24 first-try successes. Zero retries, zero waits.
Qwen3.8-Flash: upstream 429s on 16 of 20 completed tasks, 41 recorded retry cycles on those tasks alone, single-task waits up to 6 retries and several minutes. The final code suite was unreachable across eight sustained retry rounds spanning ~50 minutes — every attempt bounced — so it's marked blocked at publish time, not failed.

Fair disclosure: that's OpenRouter's shared pool, not a dedicated provider key. But if your mental model of "cheap flash model" comes from a benchmark table, this is the column that's missing: a model you can't call is 0% intelligent, whatever its DeepSWE score. Two days post-launch, demand for Qwen3.8-Flash's pool is clearly outrunning capacity. If you're wiring it into production this week, budget for backoff logic or bring your own Alibaba key.

The bill

	GLM-5.3-Flash	Qwen3.8-Flash
Tasks completed	24/24	20/24 (code suite blocked)
Total cost	$0.0115	$0.0098
Cost per task	$0.00048	$0.00049
Median successful-call latency	~5.0s	~17.2s
Retries	0	41

Flash-model API spend is a rounding error at any realistic volume. The real cost is engineering time spent verifying output quality — which is exactly what suites like this exist to compress into an afternoon.

When to pick which

Your workload	Pick	Why
Structured extraction / ETL glue	Either	Schema validity 10/10 both; disagreements are judgment calls
Hard format constraints	Either	10/10 both; token verbosity differs, invoice doesn't
Small code generation & repair	GLM-5.3-Flash (today)	4/4 recorded; Qwen's suite blocked, untested in this window
Latency-sensitive or high-volume this week	GLM-5.3-Flash	0 retries vs a 429 wall; 5s vs 17s median
Reasoning-heavy, tight token budgets	Watch Qwen	Half the output tokens for the same answers — if you can reach it

Limitations

One day, one shared pool, temperature 0, single attempt per prompt. N=24, not 2,400.
My gold answers had at least one real bug (the changelog task). Two-tier regrading exists precisely because strict grading miscounted defensible variants as errors — your graders will have the same problem.
Programmatic graders can't see creativity, tone, or long-horizon agent behavior. For agent workloads, the labs' DeepSWE/CoWorkBench numbers remain the better signal.
Qwen's availability picture is launch-week demand on a shared pool and may settle; its code-suite row is blocked, not failed.

The task set is 30 minutes of work to replicate against your own workloads — and your workload decides this better than any benchmark table. If you run it, I'd genuinely like to see your numbers.

RenoBrief - AI home redesign with contractor-ready renovation briefs

li wujie — Mon, 25 May 2026 14:27:53 +0000

RenoBrief AI: Turn Any Photo into a Renovation Plan in Seconds

Upload a room or exterior photo, get a stunning redesign — plus a budget, material list, and contractor-ready brief. No design degree needed — just snap, style, and plan.

One Tool, From Vision to Action

AI Interior & Exterior Remodel — Upload any room or house photo, choose a style and budget, get a before/after redesign instantly.
Contractor-Ready Brief — Every generation includes design direction, budget range, materials & finishes, and contractor notes. Not just a pretty picture.
Budget-Aware Design — Pick your range: Refresh under $500, Remodel under $2,000, or Premium $10,000+. AI designs within your budget, not against it.
AI Floor Plan Generator — Enter room count, square footage, and layout preferences. Get a 2D floor plan with rationale and builder notes.
Export & Share in One Click — Download a PDF brief to hand your contractor, partner, or agent. Copy as Markdown. No back-and-forth explaining what you want.

Try Now！

CleanPaste.org - Remove Invisible Characters & AI Watermarks (100% Free)

li wujie — Sun, 29 Mar 2026 14:03:47 +0000

CleanPaste.org: The #1 Free AI Watermark Remover

Instantly erase AI watermarks and hidden metadata with our powerful online tool, ensuring your text is completely undetectable and looks 100% human-written in seconds.

One Click, Every AI Trace Erased

Instantly Erase Statistical AI Watermarks
Strip Hidden Metadata & Invisible Characters
Humanize Text to Bypass AI Detectors
Ensure Originality with a Plagiarism Check

Try now!

AI Fruit - Create Viral AI Fruit Videos in Seconds

li wujie — Sat, 21 Mar 2026 05:50:52 +0000

AI Fruit: Create Viral AI Fruit Videos in Seconds

The easiest way to make fruit-eating-fruit ASMR videos for TikTok, Instagram & YouTube. No editing skills needed — just pick a fruit, choose a vibe, and let AI do the rest.

One Tool, Endless Viral Content

50+ Fruit Templates — Pre-built prompts for every trending fruit style. One click to generate.
Multiple AI Models — Powered by Wan 2.5, Seedance, Hailuo, Kling & more. Pick quality or speed.
TikTok-Ready Formats — 9:16 vertical, 16:9 landscape, 1:1 square. Watermark-free with commercial license.
Auto ASMR Sound Effects — Crunchy, juicy, satisfying audio that hooks viewers and drives views.
Videos Ready in Under 60 Seconds — Batch create content for your niche account at scale.

Try now!

City Map Poster Generator — Turn Any Location Into Printable Wall Art

li wujie — Tue, 03 Mar 2026 06:13:15 +0000

The Easiest City Map Poster Generator for Stunning Wall Art

Stop settling for generic prints or expensive custom designs. Map Poster Generator uses AI to transform any city, neighborhood, or street on Earth into beautiful, one-of-a-kind map posters. Whether it's the city where you fell in love, the hometown you grew up in, or a dream destination you've always wanted to visit, our city map poster generator turns your favorite places into high-resolution, print-ready wall art in seconds. Design faster, gift smarter, and decorate with meaning.

One Tool, Endless Map Poster Styles

AI-Powered City Map Art: Every city map poster is uniquely generated by AI — no two designs are ever the same
12+ Art Styles: Watercolor, minimalist, vintage, modern, blueprint, neon, and more to match any room or mood
Any Location Worldwide: From major capitals to small towns, generate a city map poster for any place on the planet
Print-Ready & Free to Try: Download high-resolution map posters in PNG, JPG, or PDF — your first poster is free

Try now!

Get Magic Tools - Discover the Best Software to Supercharge Your Workflow

li wujie — Tue, 27 Jan 2026 17:49:36 +0000

The Ultimate Directory for Modern Digital Creators

Stop wasting time on manual tasks or searching for the right software. Get Magic Tools connects you with the web's most powerful utilities, SaaS, and productivity apps. Whether it's cutting-edge AI, essential developer tools, or creative assets, our curated directory helps you find the "magic" solutions that solve real problems instantly. Build faster, design better, and work smarter.

One Platform, Endless Productivity

Hand-Picked Quality: Rigorously tested tools that actually provide value
Beyond Just AI: A complete collection of SaaS, dev utilities, and creative apps
Boost Efficiency: Discover software designed to automate and simplify your work
For Makers & Pros: Essential resources for founders, developers, and designers

Try now!

Foundation Men - Transform Your Look with AI Before Making Any Changes

li wujie — Mon, 12 Jan 2026 15:31:25 +0000

AI-Powered Grooming Tools for the Modern Man

Stop guessing your skin tone or wondering if a new hairstyle suits you. FoundationMen uses advanced AI to analyze your skin, simulate invisible makeup looks, and preview hairstyles—all from a single selfie. Whether you want to cover imperfections naturally or find your next haircut, our tools give you confidence before you commit. Try it free today and see the difference.

One Photo, Three Powerful Transformations

Analyze Your Skin Tone & Undertone with 99% Accuracy
Preview Invisible Makeup That No One Will Notice
Try Different Hairstyles Before Visiting the Barber
Make Confident Grooming Decisions Backed by AI

Try now!

AnyToURL - Turn Any File into a Shareable URL

li wujie — Sat, 03 Jan 2026 14:11:29 +0000

Turn Any File into a Shareable Link Instantly

Stop wasting time with clunky file transfer services. AnyToURL transforms your files into direct, shareable URLs in seconds. Whether you need to share images, videos, PDFs, or code snippets, our lightning-fast platform handles it all. Powered by a global edge CDN for blazing speeds worldwide. Sign up free today and start sharing in seconds.

One Platform, Endless Sharing Possibilities

Drag & Drop Any File, Get a Direct Link Instantly
Share Anywhere with Password Protection Options
Deliver at Lightning Speed via Global Edge CDN
Support for Images, Videos, PDFs, Code & More

Try now!

MakeBanner.ai - Professional Visuals, Zero Design Skills.

li wujie — Wed, 31 Dec 2025 00:59:00 +0000

Create Stunning Banners Instantly with AI

Stop struggling with complicated design tools. MakeBanner.ai turns your ideas into professional graphics instantly. Whether you need a striking YouTube banner, a polished LinkedIn header, or eye-catching social media posts, our #1 AI generator handles it all. Try it for free today and visualize your brand in seconds.

One Engine, Limitless Visual Possibilities

Turn Ideas into Reality with Text-to-Banner AI
Optimize Automatically for YouTube, LinkedIn & Social Media
Personalize Your Brand with Unique Professional Styles
Export in High Quality for Instant Publishing

Try now

Imgezy - The Ultimate AI Photo Editor for Effortless Creativity

li wujie — Tue, 30 Dec 2025 00:47:55 +0000

Imgezy: The Only AI Photo Editor You'll Ever Need

Turn your words into stunningly edited photos in seconds. Describe any change you want—from removing objects to changing backgrounds—and let our AI handle the rest effortlessly.

One Tool, Limitless Creativity

Edit with Simple Words Describe your desired edit, like "remove the person on the left," instead of using complex tools.
Flawless Object Removal Our AI intelligently erases unwanted objects with pixel-perfect accuracy for a clean result.
Instant Background Change Swap your background to a sunny beach or a professional studio with a simple command.
Perfect for Any Use Case Trusted by 200,000+ creators for e-commerce, travel photos, and eye-catching social media content.

Start creating with the power of AI today!

Try now!

Sketch To | AI Image to Sketch Converter

li wujie — Wed, 08 Oct 2025 03:55:30 +0000

Instantly convert your photos into art and bring your sketches to life with our powerful, free AI generator.

Key Features:

Dual AI Power: Transform any photo into a detailed sketch, or reverse the process and turn your drawings into photorealistic images.
Always Free Standard Model: Get high-quality artistic conversions perfect for social media, creative projects, and hobbies—completely free, forever.
Professional Model Upgrade: Unlock a world-class AI for unparalleled realism and detail, ideal for artists, designers, and commercial work.
Simple & Flexible: No software to install and no subscriptions required; use our free tool instantly and purchase credits only when you need the Pro model's power.

Try now

Lost in Format Conversion? A Little Tool That Saved My Day

li wujie — Mon, 14 Jul 2025 10:08:34 +0000

Hey everyone,

Lately, I've been wrestling with a pretty common problem, and I figured some of you might have experienced the same. I often stumble upon really insightful articles or useful web resources, but the format just doesn't fit my current need.

For instance, I found this amazing guide for a new library I was trying out. I wanted to take some key sections and integrate them into my notes, which are all in Markdown. Copy-pasting was a nightmare – formatting got all messed up, code blocks were a mess... you know the drill.

Another time, I needed to quickly share an entire webpage with a colleague who was offline. Sending a link wasn't ideal, and I really didn't want them to have to sift through a potentially cluttered page. A clean PDF seemed like the perfect solution.

I even had a situation where I needed to extract all the text content from a heavily designed webpage to analyze it. Stripping away all the CSS and extra elements manually was just tedious and time-consuming.

It felt like I was constantly jumping between different online tools, each specializing in one type of conversion. It was inefficient and honestly, pretty frustrating.

That's when I stumbled upon a little online tool that has become a real game-changer for me. It's called "URL to Any," and it basically does exactly what the name suggests – it takes any URL and lets you convert it to a bunch of different formats.

What I really liked about it was how simple it was. No sign-up, no complicated settings. You just paste the link, choose your desired format (Markdown, PDF, text, even images and QR codes!), and boom – you have your converted content in seconds. And the best part? It's completely free.

For someone who juggles content creation, documentation, and a bit of development work, having this one-stop solution has been incredibly helpful. It has saved me a ton of time and the headache of dealing with formatting issues.

I thought this might be useful for some of you out there who face similar challenges. Have you ever found yourself in a format conversion rabbit hole? What are your go-to solutions? I'd love to hear your experiences in the comments below!

Try it free

Let me know what you think! I tried to keep it in a story/sharing format, focusing on the problem and how the tool provided a solution without being overly salesy.