Their stats look ok, but when I tested it[0], it was 4x slower than 4.20.

[0]: https://aibenchy.com/compare/x-ai-grok-4-20-medium/x-ai-grok...

It is cheaper per token, but it seems to reason a lot more, leading to costs similar to 4.20, but performance is better (similar to what 4.20 had[0]).

Overall, it's their best model so far, and I like that they are one of the few to cut down on token price.

[0]: https://aibenchy.com/compare/x-ai-grok-4-20-medium/x-ai-grok...

Is it possible to do some sort of Binary* Search (Binary Star, as in A* star search algorithm, where we use heuristics).

    a: [1,3,5,7,8,9,10,15]  
    x: 8 (query value)

For this array, we would compare a[0], a[3], a[7] (left/mid/right) by subtracting 9.

And we would get d=[-7, -1, 7]

Now, normally, with binary search, because 8 > mid, we would go to (mid+right)/2, BUT we already have some extra information: we see that x is closer to a[3] (diff of 1) than a[7] (diff of 7), so instead of going to the middle between mid and right, we could choose a new "mid" point that's closer to the desired value (maybe as a ratio of (d[right]-d[mid]).

so left=mid+1, right stays the same, but new mid is NOT half of left and right, it is (left+right)/2 + ratioOffset

Where ratioOffset makes the mid go closer to left or right, depending on d.

The idea is quite obvious, so I am pretty sure it already exists.

But what if we use SIMD, with it? So we know not only which block the number is in in, but also, which part of the block the number is likely in. Or is this what the article actually says?

Yeah this is basically interpolation search

Oh, that's what the article was referring to with "interpolation".

Weird that I didn't hear about it before, it's not that used in practice?

One reason I could see is that binary search is fast enough and easy to implement. Even on largest datasets it's still just a few tens of loop iterations.

But with new hardware comping out, and maybe models being smart enough to help with optimizing them and reducing inference costs even more, I think we should still expect the costs to go down.

Can you use those AI cards for gaming too?

Or the makers intentionally nerf them, in order to better segment the markets/product lines?

The drivers often need per game optimisations these will be missing but I doubt Intel would nerf them, just rely on you not paying a lot for RAM the game won't use.

I actually meant it in a different way. I would get it for local AI stuff, but being able to game on it would be a huge plus, otherwise I would need two different machines.

Much as I want diversity; a 3090 would be a billion times better for games and can probably hold its own for a broader AI workload. Anything other then running highly quantised models that don't fit in 24GB with realativly small contexts.

A 3090 is what I have now.

But I hope to somehow have 48Gb or 64GB VRAM in a GPU that's also gaming-ready.

I was looking for maybe getting a mac studio for this reason, but I don't think a mac is really good for for gaming.

It'll work just fine for gaming. It's what the B770 would have been if it had 32GB RAM and ever got released.

They nerf gaming cards to make money on the pro cards. Since this is a pro card it's not nerfed.

I am a bit confused by the separation between VSCode and Copilot. If I cancel my Pro+ subscription, can I still use Copilot with my own OpenRouter key?

Yeah, but you get the benefit of using any model of your choice.

That's so cool. I would have expected for the domain to go for hundreds of thousands or millions, or, more likely, not not be purchasable for some reason. I can see a future where google.com is purchased for fun by some robot in 200 years.

Do we also have to pay for the API usage? Then they will actually be profitable, lol

Their API issues seemed to have been resolved, now it does[0] as expected, similar to GLM 5 level.

[0]: https://aibenchy.com/compare/deepseek-deepseek-v4-flash-high...