kzrdude

Go is great if you prefer deepseek V4 flash. Then it's very hard to use up the allowance.

Maybe this is relevant? Just in case

For autonomous subagents with tool calls, code execution, or multi-step reasoning: set thinking_level to "medium" or "high" to prevent premature tool termination.

I just happened to see that in the docs: https://ai.google.dev/gemini-api/docs/latest-model

He's already proven repeatedly that he'll break out of containment

It's smarter than DeepSeek v4 Pro (preview) on several benchmarks, like HLE.

Well, you can use the google models from Pi. Go to aistudio.google.com and set up an API key. There is a free quota, it's relatively large for the Flash Lite models.

...and last time I looked the limits were more generous for Gemma 4 there, but they have been tightened a bit. That's how it goes, always changing.

I asked Gemma 4 E2B, and if you use it as a reasoning model, it will give a better answer (that it doesn't know; it was also using the date information from the prompt.)

Google's models are always very well spoken and much more pleasant to talk to. And I've used the open weight models a lot, too.

I've never questioned the word lite before because it's existed my whole life.. So does it make sense? Why does it exist and where does it come from? More than coming from "light".

That sounds smart, maybe worth testing, but I think it bloats the instruction cache even when not tracing.

If you go small enough it should be no problem. For example Gemma 4 E4B in Q6 or Q4 quantization should run well on your laptop. It shouldn't be too taxing, but would still want to eat 7-9 GB of VRAM or so.

Now that model is mostly useful for writing or chatting.

LM Studio seems to do the same thing, except that LM Studio is not open source. So they have a point, they do something more.

Looks like Portuguese to me (I don't speak it)

Wikipedia has policies and it needs to use reliable sources. This rule is often skirted and a lot of facts are cited to self-published and fast moving web sources. Those who are braking the progress on the article here are doing the right thing, trying to uphold the editorial standard.

Luckily there is now a New Scientist article to link to, so, the issue should now be resolved.

Which exact model does claude use here?

Mario Zechner's talk "Building Pi in a World of Slop" is a good listen, on the topic of adults in a "world of slop". https://www.youtube.com/watch?v=RjfbvDXpFls

Kimi K3 is on the opencode-go plan's price list now. I haven't confirmed it, but I guess that means it's available. (Not in Zen.)

Fwiw, American industry has given away a lot for free - you could include large parts of the open source movement in that - and all the "free" VC backed services like facebook would be another prong of the same comparison. I would rather compare this way, that China is gaining soft power and goodwill, in the technology and innovation sense, in a way that's similar to how USA has done in the past.

If DeepSeek v4 flash is run using Q2, then people should run this one using Q½ or maybe Q¼

OpenRouter's ToS also seems to allow them to store your submitted prompts anyway, so privacy advocates would have to look elsewhere anyway, that's at least how I understand it (and it surprised me).

Google has released some very good and useful models too, Gemma, but they are not so large. But they are in useful size.

What is it useful for at slow speed?

Have defined behaviour on any input, i.e not causing security issues, memory safety issues on untrusted inputs. Lz4 has this.

ChatGPT is a household name. And OpenAi is actually not, people outside tech don't necessarily know it.

I didn't know that Amish thought so lowly of their own language, I think that's just sad. It's their own language and there's no reason to measure it against others.

Someone posted this fork, which should work in just 10 GB RAM, at least for another big model. I thought of your request and came back to spread the news :)

https://github.com/ErikTromp/colibri-hy3

How many tokens/s do people feel are enough? Even 5 tok/s would be hard to use in an interesting way, not sure what the use is.

It might be impossible to run such a large model on the laptop, if I were you I'd first start experimenting with running smaller models, using llama.cpp or lmstudio, just to start understanding how it works and learn about some of the concepts involved - quantization and all!

For example Qwen3.5-4B or Qwen3.5-9B could probably be run cpu-only on some quantizations at more reasonable speeds.

If you want a larger one, you could probably run Qwen3.6-27B or 35B-A3B decently slowly.

Now to be humble - maybe it's impossible to run colibri (this project in the OP) on your computer. Note the performance stories - 0.07 token/s on a 24 GB laptop with 24 CPU threads!

When the weights are publically available (and open to use), then there's a free market for hosting that model at least. That's not independence for me, but it's lack of vendor lock-in.

For some of the open models, there's a list of 20-30 providers of the same model on openrouter for example, as an example of the supply.

Yeah but it also seems like they have the best model right now?

It aligns pretty well with corporate and pr-speak, but the LLMs really weaponize it and tire out the phrases quickly. I bet this is interesting for linguists.