# Qwen3-VL

51 Hacker News commenters recommended Qwen3-VL and 10 criticized it, November 2025 to November 2026. It was recommended in 12 of 13 months. It meets the Hacker Approved badge rule.

**Verdict:** Open vision-language model run locally for OCR and image or video understanding, praised for speed. Critics say it misses small details and gets click coordinates wrong.

## Why people recommend it

- Strong OCR and text reading (3 comments): "i use it almost exclusively to OCR handwritten todo lists into my todo app and i don't think it's missed yet, does a great job of toolcalling everything" ([source](https://news.ycombinator.com/item?id=46008874)) "has made translating screenshots of text so much easier via LM Studio" ([source](https://news.ycombinator.com/item?id=45883951)) "It is much much better than my experience with Tesseract in general." ([source](https://news.ycombinator.com/item?id=49421214))
- Fast local performance on modest hardware (4 comments): "it leaves me plenty of RAM for workstation tasks while delivering tokens at around 30tok/s" ([source](https://news.ycombinator.com/item?id=46915397)) "use qwen-30b-vl all the time locally as my goto model because it's just so insanely fast" ([source](https://news.ycombinator.com/item?id=46095133)) "my meager M1 Pro is able to keep up without the fans spinning 100% of the time" ([source](https://news.ycombinator.com/item?id=47055987))
- Good spatial grounding and image understanding (4 comments): "They manage to capture spatial relationships, objects, and semantics from the image, it's very impressive." ([source](https://news.ycombinator.com/item?id=46133229)) "Qwen was the only model consistent enough for production—good spatial awareness, reliable JSON" ([source](https://news.ycombinator.com/item?id=46182653)) "Qwen3VL models are smaller/faster and better "spatially grounded" in pixel space" ([source](https://news.ycombinator.com/item?id=46766675))
- Beats larger and proprietary rivals (4 comments): "The 9x smaller model outperforms the larger one from earlier the same year on 27 of the 40 benchmarks" ([source](https://news.ycombinator.com/item?id=46552512)) "It definitely has outperformed Gemini which generally is much better than openai or Claude on videos." ([source](https://news.ycombinator.com/item?id=46127660)) "Qwen VL models nail tasks frontier models dont even get close to acceptable on." ([source](https://news.ycombinator.com/item?id=48990462))

## What people warn about

- Misses small details or gets values wrong (3 comments): "unfortunately they can miss the small details and didn't provide as much value as I was hoping" ([source](https://news.ycombinator.com/item?id=47210214)) "It mixes up the rows, resulting in many values associated with the wrong substance." ([source](https://news.ycombinator.com/item?id=46044660)) "Failed miserably, although I was using the 30B-A3B model instead of the larger one." ([source](https://news.ycombinator.com/item?id=46129328))
- Inaccurate pixel coordinates for clicking (2 comments): "it knows what to click but struggles to accurately return which pixel coordinates to click" ([source](https://news.ycombinator.com/item?id=46130026)) "It's very not accurate, but sometimes instructing to return pyautogui code works." ([source](https://news.ycombinator.com/item?id=46130339))
- Not production-ready for many tasks (2 comments): "Not production-ready but the open-source vision models are getting surprisingly good at this." ([source](https://news.ycombinator.com/item?id=48519800)) "Still not great at the use cases I tested it for" ([source](https://news.ycombinator.com/item?id=46130464))
- Sycophantic, cliched or evasive answers (2 comments): "although it was not perfect (sycophancy, overuse of certain cliches...nothing I can't get rid of later with some custom promptsets" ([source](https://news.ycombinator.com/item?id=46104137)) "it is evidently trained to be careful and not answer too specifically about that topic" ([source](https://news.ycombinator.com/item?id=46769375))

## Cite as

Hacker Approved, "Qwen3-VL", https://hackerapproved.com/p/qwen3-vl/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).