Qwen3-VL

“I believe those models have video encoders can better perform at this scenario.”

“recommended the best replacement small model (Qwen VL 3B something)”

“It is much much better than my experience with Tesseract in general.”

“even Qwen3VL 8B does an excellent job at this, despite being quite old”

“Qwen VL models nail tasks frontier models dont even get close to acceptable on.”

“Not production-ready but the open-source vision models are getting surprisingly good at this.”

“qwen model is 1/3 of the size, runs faster, and is far more reliable”

“Qwen3.5VL is awesome and it's training vocab is quiet strong”

“A file watcher on my NAS sends new images to it, which autocaptions and adds the text descriptions”

“Gemma3:27b and Qwen3-vl:30b-a3b are among my favorite local LLMs”

“unfortunately they can miss the small details and didn't provide as much value as I was hoping”

“my meager M1 Pro is able to keep up without the fans spinning 100% of the time”

“it leaves me plenty of RAM for workstation tasks while delivering tokens at around 30tok/s”

“it works reasonably well with something like qwen/qwen3-vl-30b (fast)”

“it is evidently trained to be careful and not answer too specifically about that topic”

“Qwen3VL models are smaller/faster and better "spatially grounded" in pixel space”

“Thanks to Qwen VL, I shared my grandmother’s recipes on the web.”

“The 9x smaller model outperforms the larger one from earlier the same year on 27 of the 40 benchmarks”

“qwen2.5vl:7b works well (in my use cases better than qwen3vl)”

“I recently read that someone got significantly better results by using F16 or even F32 versions of the vision model part”

“I've found them to be fantastic as this kind of thing so far”

“the "reasoning over image data" progress with things like Qwen3-VL is very impressive”

“I've found Qwen3-VL to be fairly accurate at detection (though it doesn't always catch every instance).”

“Qwen was the only model consistent enough for production—good spatial awareness, reliable JSON”

“Performs great, however god forbid you try to get that much memory these days.”

“They manage to capture spatial relationships, objects, and semantics from the image, it's very impressive.”

“In my experience, Qwen3-VL models have very accurate grounding capabilities”

“Still not great at the use cases I tested it for”

“It's very not accurate, but sometimes instructing to return pyautogui code works.”

“it knows what to click but struggles to accurately return which pixel coordinates to click”

“Failed miserably, although I was using the 30B-A3B model instead of the larger one.”

“It definitely has outperformed Gemini which generally is much better than openai or Claude on videos.”

“although it was not perfect (sycophancy, overuse of certain cliches...nothing I can't get rid of later with some custom promptsets”

“use qwen-30b-vl all the time locally as my goto model because it's just so insanely fast”

“the LLM does "reasonably well", in that it accelerates finding the right ways to wrangle a page with Playwright”

“It mixes up the rows, resulting in many values associated with the wrong substance.”

“the 8B of the Qwen3-VL family 9th place, above a few proprietary models”

“it is quicker to use it to answer most basic google questions”

“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”

“has made translating screenshots of text so much easier via LM Studio”