Comment by therealpygon

“Justified or not” — is certainly a useful caveat when giving the same credit to a few people who complain loudly with mostly unauthentic complaints.

> Vendor lock-in

That is, probably the most ridiculous of the statements. Ollama is open source, llama.cpp is open source, llamafiles are zip files that contain quantized versions of models openly available to be run with numerous other providers. Their llama.cpp changes are primarily for performance and compatibility. Yes, they run a registry on ollama.com for pre-packed, pre-quantized versions of models that are, again, openly available.

> Closed-source tweaks

Oh so many things wrong in a short sentence. Llama.cpp is MIT licensed, not GPL license. A proprietary fork is perfectly legitimate use. Also.. “proprietary“? The source code is literally available, including the patches, on GitHub in ollama/ollama project, in the “llama” folder with a patch file as recent as yesterday?

> Mixed Performance

Yes, almost anything suffers degraded performance when the goal is usability instead of performance. It is why people use C# instead of Assembly or punch cards. Performance isn’t the only metric, which makes this a useless point.

> Opaque model name

Sure, their official models have some ambiguities sometimes. I don’t know know that is the “problem” that people make it out to be when ollama is designed for average people to run models, and so a decision like “ollama run qwen3” not being the absolutely maximum best option possible rather than the option most people can run makes sense. Do really think it is advantageous or user friendly, when Tommy wants to try out “Deepseek-r1” on his potato laptop that a 671b parameter model too large to fit on almost anything consumer computer is the right choice and that it is instead meant as a “deception”? That seems…disingenuous. Not to mention, they are clearly listed as such on ollama.com, where in black and white it says the deep seek-r1 by default refers with the qwen model, and that the full model is available as deep seek-r1:671b

> Context Window

Probably the only fair and legitimate criticism of your entire comment.

I’m not an ollama defender or champion, couldn’t care about the company, and I barely use ollama (mostly just to run qwen3-8b for embedding). It really is just that most of these complaints you’re sharing from others seem to have TikTok-level fact checking.