HN user
They basically don't exist in the currently most profitable LLM market (coding).
Yes, subs like codex are heavily subsidized. But API billing has massive margins and that's what enterprises pay.
It was the agents job to pick the model.
Running above the display refresh rate is only decent when you are wayyyy above it, at least double. If you have a 120hz display and run something variable around 150fps, its much worse. But 500fps does help, you're basically seeing more recent frames as it scans the screen.
Following rules like "verify all changes with tests" down to a tee is usually a desirable trait in LLMs. Personally I'd leave that behavior there (just like with humans for some tasks like aviation you have them go through checklists even if some stuff you can infer is not needed). But otherwise just make it "always run tests unless you're absolutely sure they can be skipped".
Same in Argentina. It's almost surely a region whitelist for now (it's the only reason Argentina ever gets blocked).
Very strong pricing, cheaper than Grok 4.5, particularly the cached reads. We'll have to wait to see if it's actually worth using (it's not on OpenRouter yet).
You can buy the Grok plan, Cursor also has a plan which includes grok 4.5, but I don't know how subsidized they are compared to codex or claude code plans.
GLM 5.2 is around 1:16.
Sounds like an issue with your harness or something.
Completely irrelevant, which was expected considering their previous models were vastly outclassed by other models at SWE.
This is the first grok model that seems actually pretty competitive at SWE.
I could easily see Grok 4.5 being around 1:16 in terms of active parameters, so around 94B active parameters.
First impressions:
- Very fast, easily beats GPT 5.5/Opus 4.8/GLM 5.2 because of higher t/s (around 90?) and very high token efficiency
- Very good price, no contest vs GPT and Opus which are very overpriced if you pay API costs, and probably cheaper than GLM 5.2 when you take into account the token efficiency.
- Will take quite a while to get a feel for how smart it is, but it's definitely good, I'd say in the same tier as opus, occupying the lower end of that tier together with GLM 5.2.
Definitely in the right direction in terms of architecture. However those "hmmm" "uh huh" interjected in the demo are pretty awful.
It's cheaper if you pay API prices. If you pay a gpt sub then codex is much much cheaper.
Yeah, it's a much better idea to buy many used 3090s. 4090s or 5090s if you can afford it. Way faster.
I bought 2 used 3090s some years ago for $500 each. They're probably a bit more expensive now, but I guess for something like $2000 you can build a barebones 2x3090 PC which will be way faster than a Macbook. (you're fine with very basic hardware outside the GPUs)
And codex is even more subsidized. It's an absurdly good deal.
That only affects people in California. Whereas Fable being shut down affects people all over the world.
These NVIDIA GPUs aren't general purpose in the way that you think. They can't even run games. Nvidia blackwell is probably slightly more efficient than TPUs for training. Do you really expect a 4 trillion company with the majority of its revenue being AI for some years now, not to have built its flagship product fully around AI? The GPU name stuck around, but they are pretty terrible at graphics.
The real efficiency win in these chips is that they are made for inference only. You can throw away the vast majority of a chip if you only need a few ops, a single precision (like INT8 or FP8) and don't need ultra fast interconnects.
That's crazy good for $2400.
I'm just saying it doesn't deserve such praise for being "real", because if you placed a spycam in a random steam user house it wouldn't look like that. Not that there's anything wrong with it being carefully crafted like any marketing material.
About 0.3% of people are firefighters in the US. The odds that you end up with 100 of them is:
0.000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000001%
That's 248 zeroes after the decimal point.
Probably not at all, considering weights are randomly initialized.
If the Anthropic, Google, and Reflection AI deals stay afloat, the numbers are much better.
If you sampled 100 steam players at random, it would look nothing like that.
That's the CLI, not the codex app which is proprietary.
I think anthropic needs that compute. They'll probably rent it as long as they can. They aren't really in the datacenter business. Google yes, I think they'll stop at some point.
If they can keep google and anthropic deals, they're already profitable.
Can you really say you're running GLM 5.2 if its a 2 bit quant? It might be usable but the capabilities will definitely not be the same.
I've been running some test prompts comparing frontier models for webdev, particularly pretty visualizations, physics / orbital simulations, etc.
Do note that GLM is not multi modal, which can be a deal breaker. And these open models are not good outside coding.