9 ms·

I dont get why they'd make this available to free users when they're probably dealing with compute constraints considering their competition with the Chinese mo

I dont get why they'd make this available to free users when they're probably dealing with compute constraints considering their competition with the Chinese models. Giving a more capable model to a huge free user base looks like an expensive choice to me

freakynit 1mo ago
They've probbaly have realized that the foundation models are becoming commoditized (or will be very soon). We can already run luna medium class models on phones now (bonsai, gemma4). The value is shifting up the stack... make core intelligence free, then monetize the ecosystem built on top of it .. kinda similar to how the internet itself is free, but platforms and apps capture the value. I think we'll see a huge push towards connectors for work and personal tools, along with much deeper os level integration. That's where the long-term moat is, not the base model itself.
judge2020 1mo ago
IMO Competition with chinese models frees up higher-turn and bigger-context tasks, since SWEs are more likely to use model routers for coding. Thinking may be a few more steps but the problems are a lot less complex and probably 1/10th the context size, even when skills and memory are at play.
dawnerd 1mo ago
For the ad integration, obviously. Plus you tell people they can good up their Apple health and now you get more free training data.
ssl-3 1mo ago
Even though it seems implicit that all of these companies are losing money hand-over-fist, it's still a competitive market. Improving the free offerings is meant to increase visibility and therefore market share. It's not about goodwill, and it never will be. :) The free stuff is primarily marketing and marketing always has costs. In terms of compute, I have no way to really look behind the curtain and see what goes on back there. But I know with codex CLI, in terms of weekly quota: I can get a ton of work done with luna and usually get reasonable results. It feels very compute-light in this way. I'm amazed by the work luna on xhigh can do for the price I pay (just $20, every month). It has the presentation of something that is very efficient to run, while also being something that can actually produce OK results. It's also fairly quick. It differs from many previous smaller offerings of yore in this way. Like, I mean: I found stuff like the -mini models and 4o to be utterly useless wastes of my time. Luna isn't like that at all; it can get some stuff done. So far for me, luna is the most impressive part of the 5.6 rollout. Not because it is best, but because it is useful and cheap. So if luna is decent (it seems that it is), and if it is in fact light (which seems to be true from what I can observe), and it is offered for free, then it may very well be better, faster, and cheaper than the competition is. And that's good for visibility. Marketing is all about buying eyeballs.
sumedh 1mo ago
The same reason Google Search is free.

freakynit 1mo ago

They've probbaly have realized that the foundation models are becoming commoditized (or will be very soon). We can already run luna medium class models on phones now (bonsai, gemma4). The value is shifting up the stack... make core intelligence free, then monetize the ecosystem built on top of it .. kinda similar to how the internet itself is free, but platforms and apps capture the value. I think we'll see a huge push towards connectors for work and personal tools, along with much deeper os level integration. That's where the long-term moat is, not the base model itself.

judge2020 1mo ago

IMO Competition with chinese models frees up higher-turn and bigger-context tasks, since SWEs are more likely to use model routers for coding. Thinking may be a few more steps but the problems are a lot less complex and probably 1/10th the context size, even when skills and memory are at play.

dawnerd 1mo ago

For the ad integration, obviously. Plus you tell people they can good up their Apple health and now you get more free training data.

ssl-3 1mo ago

Even though it seems implicit that all of these companies are losing money hand-over-fist, it's still a competitive market. Improving the free offerings is meant to increase visibility and therefore market share. It's not about goodwill, and it never will be. :) The free stuff is primarily marketing and marketing always has costs. In terms of compute, I have no way to really look behind the curtain and see what goes on back there. But I know with codex CLI, in terms of weekly quota: I can get a ton of work done with luna and usually get reasonable results. It feels very compute-light in this way. I'm amazed by the work luna on xhigh can do for the price I pay (just $20, every month). It has the presentation of something that is very efficient to run, while also being something that can actually produce OK results. It's also fairly quick. It differs from many previous smaller offerings of yore in this way. Like, I mean: I found stuff like the -mini models and 4o to be utterly useless wastes of my time. Luna isn't like that at all; it can get some stuff done. So far for me, luna is the most impressive part of the 5.6 rollout. Not because it is best, but because it is useful and cheap. So if luna is decent (it seems that it is), and if it is in fact light (which seems to be true from what I can observe), and it is offered for free, then it may very well be better, faster, and cheaper than the competition is. And that's good for visibility. Marketing is all about buying eyeballs.

sumedh 1mo ago

The same reason Google Search is free.