Why the gatekeeping? It’s their choice how they spend their time and tokens, and nobody is forced to use it.

As long as they had fun and enjoyed the process, that’s what counts. Enterprise software deployment is a different story, of course.

Buy the dip. (Sort of joking?)

Significantly slower and less capable than opus 5.5

However you can run comparable or better models than gpt4. Try to stay under 30b

GPT-4 and Opus 4.8 were the models that started the AI craze. Now people look down on them.

Qwen 3.8 27b is a great model, but 2 bit quant of it is _not_ Opus 4.8 level.

WER is not good enough for professional use, fine for hobby projects

I’ll just use open source, thanks

Why are you using Ollama? Just use llama.cpp

MLX?

Then use oMLX or MTPLX or Rapid MLX.

This is very fair. Compared to Claude the Codex limits are much better and smarter model.

Just a reminder: The Netherlands' 612 tonnes represent 0.28% of all gold ever mined and 1.7% of central bank reserves. Because gold is insanely dense (19.32 g/cm³), all the gold mined in human history fits into about ~340 shipping containers

A standard 24,000-TEU cargo ship could hold every ounce of humanity's gold using under 1.5% of its physical volume.

So the Dutch portion is a rounding error of global gold that fits in a single shipping container.

That shipping container could not be easily lifted, right? The bottom would simply rip off under so much weight.

True and the same for the lorry or train. Likely 30 needed but they’d be very empty.

If it’s anything like Reachy, can run it fully offline inc inference