redox99

It's very likely tencent games those stats, buying their own tokens.

Because using chinese models on OpenRouter is way more expensive than the equivalent codex sub, unless you use something clearly inferior like deepseek flash.

Codex sub is way cheaper than equally strong models on openrouter.

> if that datacenter deal they signed with Musk changes anything

It's hard to imagine a 1B / month deal wouldn't.

It's always been clear to me that Cursor's business model of reselling others' APIs had its days numbered. Not necessarily because the providers would pull the plug, but because you wouldn't be able to compete with subsidized plans.

Cursor is already kind of useless for third party models unless you're willing to spend thousands of dollars. It's only worth it if you're going to use mostly grok/composer.

I find GLM 5.3 Flash more interesting than 5.3. The fact 5.3 does not have vision is kind of a deal breaker. Also 5.3 Flash seems to be better at making pretty UIs.

Because obviously everyone who participates in a discussion has an account, so they don't feel this annoyance other than maybe when sharing a link to someone else.

Yeah, exact same for me. K3 used to be my go to for UI but it's quite expensive. Ox alpha being so cheap and so comparatively good at design is crazy.

I haven't tried it but I think Qwen Max is also very good at design.

Yeah, when it was secretly called Ox Alpha.

Search results suck because the web sucks these days. The big models from OpenAI/Anthropic have every book in existence baked into them

At 125B + 51B I'd expect it to have some degree of world knowledge, clearly in the middle between small models like qwen 27B, and huge trillion parameter models.

It's also better than Sol (at whatever effort) at designing pretty UIs. I have a Codex sub and I've been using this model for UI stuff.

Sure, although I still expect it to become the most used model on openrouter.

Not really a brag: it ran like shit. Very slow (~20tps, VERY high latency) and it would timeout all the time.

I'm sure the chips are fine, but they clearly didn't have enough capacity for the demand they had (that 100T/day claim was asbolute bs)

Ox alpha is better at UI than GPT 5.6 Sol. Not a high bar considering Sol sucks at UI, but as someone who just has a codex sub, I've used almost 1B tokens of ox alpha these last few days to complement Sol smartness.

Inference was atrocious in terms of speed and constant timeouts. If it's served fast it will be a delight to use.

Ox alpha at moments felt like it was quantized to hell. I think the last few days it might have improved.

But everyone can load 10 bucks on a deepseek flash API and have almost unlimited usage at decent speed.

That'd be ungodly expensive.

Both are still very low for AI btw.

The distribution in those offices is probably the same, a small percentage of people cause almost all the nastiness.

Sol is much cheaper than opus because it uses less tokens. Never compare token pricing.

It's a mistake to think only OpenAI and Anthropic are actually spending the big bucks on pretrain, and the others just distill that.

The Chinese models are pretrained on large clusters just like OpenAI ones are. Yes, they use outputs of the frontier models to further improve the final model, but even without those outputs they'd still have very strong models.

It's not like in a world without distillation things would be much different as you claim.

> i'm not sure why people expect agents to one shot everything to perfection with just a prompt.

because that's the end goal? and for simple small stuff they're already there?

Yeah if you're unlucky you get hit with like 20 years for wire fraud.

I used laptop because that's what the parent said, but I really meant a computer of any kind at home. I assumed a remote interview. If its an interview in person at an office I'd definitely expect hardware to be provided.

I wouldn't want to hire someone who literally never uses a computer outside work (to the point of not owning one).

Then you should find a way around it. If an applicant can't even manage to get a laptop for a few hours I rather hire someone else.

In extreme cases we can work it out but you should have something that makes up for the extra annoyance compared to 99% of applicants.

I have no problem running two or three sequences of qwen 27B with a 3090. It's basically the recommended way, LLM inference without batching is super inefficient.

Most people probably don't care about nginx performance. It shouldn't be your bottleneck unless you serve massive amounts of static data.

You don't really need to train a 10T model to test cerebras against a 10T model. You can feed it an untrained (randomly initialized) model and benchmark it. Result will be gibberish but performance the same.