GLM-4.6V

[RichardScottOZ]

I have been looking at all the vlms on openrouter and their native capabilities - for comics, this model is very impressive.

Gemini models are the clear leader here, but GLM-4.6V approach Gemini 3.0 Flash - there is no Flash Lite. This model is cheaper in testing of 1000 or so pages - but wordier in tokens, so probably like 70% the cost at 0.30/0.90.

It can one shot grounding bounding boxes that are better than faster rcnn generally with no problem.

[RichardScottOZ]

GLM 4.6V Flash which I discovered yesterday.

https://huggingface.co/bartowski/zai-org_GLM-4.6V-Flash-GGUF/discussions/2

or https://z.ai/manage-apikey/rate-limits - the api is currently free. So pretty good value.

It also one-shot good bounding boxes - so this is BETTER than gemini 2.5 flash from a 9B model.

I asked the z.ai chat if it had been trained on comics and it said yes.

[RichardScottOZ]

The parent version can do it - but this is a 9B model Flash first try.

Guardian page 003.

[RichardScottOZ]

https://github.com/zai-org/GLM-V