I don’t think they will until they change the architecture.

They don’t have a prefix cache like other providers, or at least don’t have a discount in their billing structure. Each message charges for the whole context window. It’s wildly more expensive for long multi turn scenarios with lots of tool calls (coding). It’s better for short few turn tasks.

Edit: I don’t know if they actually have a proper cache. This could just be a billing artifact.

> GPT-OSS 120B which is nigh useless nowadays:

I still think that was a really great model that got overlooked. It was really great in terms of latency/throughput while still being fairly intelligent.

I was planning on using it for a design tool, but moved over to luna since it's comparable speeds and cost for a lot more intelligence.

You cannot infer this because they only show the tokens per second per user. One way to get a higher number is to have fewer users per chip.

I'm pretty sure Cerebras has a confidentiality agreement with OpenAI, and this press release was carefully constructed to avoid leaking details about the model weights. For example, the graph of tokens per second vs. tokens per second per user doesn't have any numbers that would allow you to translate between the two. (And in any case the relationship depends on the model.)

Didn't they say that they can support bigger models now?

(Where did you see that?)

This was also interesting: "CS-4 delivers more than 1,000 tokens per second on models exceeding 10 trillion parameters." Was it known that there were 10 trillion parameter models in use?

I think the frontier providers keep the size of their models carefully hidden.

Like NVIDIA bought Groq, AMD might do well buying Cerebras.

Maybe, but GPU is just one aspect of NVIDIA's dominance. If you are buying Vera Rubin GPUs, you're getting an NVL72 rack, which is only one of several racks that you're probably buying. You'll also need your NVIDIA racks with NVIDIA networking & storage gear, too. At the end of the day, they're "vertically integrated" for your accelerated computing data center (e.g. the "AI Factory"). This doesn't even count the software layer, where CUDA + CUDA-X (not to mention the software for all the sysadmin pieces) has a huge first mover advantage over anyone else.

Press doubt. Single GPU? Maybe. MultiGPU behemoths like NVL144 and NVL576? I don't think so.

NVLink is at gen9. they had a lot of teething problems and can codesign the hardware and software.

in the name of openness (AMD's only """weapon"""), the UALink spec is a hodgepodge of corporate opinions with very different implementations (looking at you, Broadcom). at spec version 1 (in hardware).

I wish them good luck as I really like AMD, but they compete no more on this than Lambo vs Bugatti.

It's not really a far fetched prediction: high margins and huge market attract competition, that's just the law of economics.

Sol is supposed to be 5T according to rumour. The imminent Astra is allegedly 10

I mean we kinda know the frontier models are multi trillion parameter models. The only open weights that are close to the frontier are that size too

cerebras model are different size then the original models

Without having any inside information, one possible theory:

All or a vast majority of of the cerebras manufacturing capacity was going to a few companies that aren't publicly available inference providers on openrouter, for their own internal use.

or

The asking price of the S-3, no matter how speedy it might be, for small/medium size customers made it economically prohibitive to purchase and use to sell public inference vs. buying more common nvidia b200 or whatever.

Cerebras provides high-speed inference at high cost. It's never going to be the cheapest and thus it will probably remain niche.

Probably the same reason why there are more people who takes buses, subways, trains than drive Ferraris.

If you're willing to pay a significant premium for latency, why use openrouter? And anyway Cerebras only supported a few specific models.

Cerebras capacity was pretty much entirely bought out at some point. We needed it and couldn't get it.

The WSE is very expensive to build, and they have a waiting list of customers who are already willing to pay a lot of money for the available supply.

it only takes ~445 GB300 NVL72 (about $22b) to run ALL of openrouter demand for a year. Microsoft rolled out $32b of DC 2026Q1.

imo the issue is that most openrouter demand is inauthentic activity (things that anthropic and openai models will refuse to do like pretend to not be bots when interacting with humans)

I use Cerebras via OpenRouter. It’s every bit as fast and reliable for my needs as claimed. I suspect the reason is that they can either be making peanuts selling inference to plebs like me via OpenRouter, or making bank selling the more expensive models to businesses directly. In short: I would be very surprised if they have die capacity, and are at this point maximising revenue per chip.

They do offer API services to individual users... though with a set of models that makes it unlikely that you want to use it. They are promising Qwen 3.8 27B any day now though*.

if you have the money as an "individual user" to purchase one of their racks... save your money and retire.

* Actually they sent out an email claiming they already have it, but I don't seem to have access, they're promising to release it to the "shared tier" any day now.

needs an sla that says power will never ever ever go out or else you will have a useless shattered plate of silicon.

I’ll get that 250kW home power service dropped in next week!

I'd like to see a consumer version too, I don't need a whole rack of them. I probably can't even afford one gpu-sized one

162 kW

"10x more throughput per watt than CS-3"