their full talk is here https://www.youtube.com/watch?v=Ok6QdZHCcI8

30 cents per million tokens is wild.

The relevant question is how easy is to setup 56x B200s in a cluster, are we talking about a adhoc infra team job or clicking a button on AWS?

Sometimes the infra team job feels easier…

Is $0.30 per million tokens that wild? Haiku 5.5 is $0.10 per million input and $0.50 per million output. And that's before you get onto the cheaper cache pricing. Surely a self hosted GLM isn't better than Haiku 5.5 at coding?

> Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI.

So they've invested around $2.5M-3M into hardware + monthly operating costs (electricity, updates, cooling). I wonder how much did they spend on tokens to decide to make such a switch. I mean if they used to spend $500k-1M per month just on tokens then it would pay off in a year or so, assuming their infra will be stable and efficient.

We'll see if that's a trend for mid-big companies or just an experiment.

I doubt local will be all that common aside from regulatory requirements but buying them off your cloud provider will be the new norm for many enterprises.

The slice of the pie where frontier models are required is just gonna be smaller and smaller. And at point where AI wrote docs, tests, and all the easy and medium difficulty functions you are already saving a ton per employee

Personally, I think Anthropic and OpenAI will introduce enterprise plans such as $2,000/month/engineer plans that will give 100x more tokens than the current max plans and then charge API pricing.

Yes, margins will go down a bit but the frontier labs still have some leverage.

The B200 was launched in 2024 if I'm not mistaken. The big gamble with this might be how fast these will deprecate in value/use as soon as Nvidia launches a new and better product.

At the moment at least Nvidia doesn't have a incentive to release a new product because it's sold out as it is and probably doesn't want to plunge their customers in depth with outdated products?

They do want to launch new product - Vera Rubin

In related news, Core 4 Solutions listed a decommissioned $60 million AI supercomputer cluster featuring 9,600 Nvidia A100 GPUs on eBay. It's six years old. Nvidia isn't Intel with tick and tock, but they're still going to come out with newer hardware every year.

The gamble in this case is pocket change. They could comfortably spend this amount every year and just use precious years for overnight / batch processing jobs when they become less useful for dev work but remain operational.

> So they've invested around $2.5M-3M into hardware + monthly operating costs (electricity, updates, cooling).

Are you assuming that they bought the hardware? because they could also be renting the metal off a provider, in which case they either pay by the hour or have a yearly commitment to save on costs.

It says in the video that they run them on Together.ai

yeah, seems like they're renting (according to the info from the video), but doesn't it make the bill worse? I mean the provider makes money somehow, so it should be more expensive for revolut. I guess it's a lot anyway

Maintenance, capacity management and spares all add up, not to mention labour. If they have a requirement to maintain geographical diversity then that moves the needle significantly if they’re primarily based on the other side of the country.

I've quickly checked and guestimate of the rental would be ~250-300k per month for this kind of usage they claim to have. It's all based on open prices from together.ai. So probably it worth the switch for them afterall, I can imagine that 56 engineers can burn tokens for more than 300k per month

Unless you have very strict requirements, it does not make any sense to invest in hardware. Most companies just rent it. Also, it's pretty hard to get the hardware.

It is in the article: "the gains showed up when tokens stopped being rationed" - so it is not just a "cost per token before/after" calculation. Once you told your employees "token are free, flatrate for all - go nuts!" they started to use it for all sorts of things and increased productivity. That is not easy to calculate.

If investors knew how powerful open models are the AI companies valuation would tank overnight

Well, with the recent news about OpenAI being unable to find investors, I feel like it's already happening

Don’t trust the news, especially now that AI is seen as part of US national security. We’re in a war, both financially and technologically.

You still need to run them on GPUs

AI companies aren't valued on the GPUs they own.

I think he meant NVIDIA is the winner in the end.

And GPUs anyone can buy.

And if the market teaches us one thing, it is that GPU prices will drop over time. The current absurd prices is attracting competitors, and the more competitors, the lower the price.

I would not be surprised if, in 5 years, there's a neat, low-power GPU in high end laptops, alternatively, a cheap, say 500 USD or so, GPU in a box, that you can use for your AI-center at home or in the office.

MoE models can be largely offloaded to system RAM - recent versions of llama.cpp automatically distribute parts of the model, with all the dense parts going into VRAM and just the routed experts residing in system RAM.

I'm currently running Qwen3 Coder Next, which is a 80B A3B model, at 15tps on a 8GB 7700S and 64GB RAM.

It's an idiot (no Thinking), so it has to go over its work several times to get it right, but it's the largest idiot I can feasibly run at the moment.

Also by accident I'm undercutting the only provider of this model on OpenRouter by 40%, doing $0.48/mTok.

Of course at my token rate it would take an entire day to even reach a million tokens, but it's the principle that matters.

No, it would be hyped up even more for a pump & dump :)

Many companies are figuring this out.

This. I don't know why OpenAI, Anthropic and others have these high valuations. We have a temporary demand shock on hardware (GPUs, RAM, Storage) but the market dynamics will create more supply in the next years. Once everybody and their mother can get their hands on hardware, we will see tons of token-per-dollar inference services competing against each others. None of these providers will do any training, just hosting open-weights model. Why would anyone give OpenAI and Anthropic their subscription money then.

It's probably true that open weight models are "good enough" for most software engineering tasks.

But the goal posts will move and I think frontier labs will find new areas that have a need for absolute frontier.

"Frontier labs" makes it sound like there's some special sauce to these companies. There isn't, other than very smart people working on problems.

Guess who else has that? The Chinese companies developing open weight models. In fact, the vast majority of AI science comes from China. They might still be playing catch up in some areas, but they're also leading in others (famously DeepSeek's inference architecture as an example).

There is IMO a non-trivial chance those labs will eventually surpass the US ones and become themselves "frontier", and that may happen sooner than a lot of people expect.

Of course, it still remains to be seen if they will keep publishing the weights at that point.

Chinese labs are gated by lack of compute, at least until they can build their own EUV machines.

At some point, they'll stop releasing models for free and will need a license.

> At some point, they'll stop releasing models for free and will need a license.

Is this a guess, or do you have any information to back that inference (no pun intended)?

It's my long standing view.

There are already rumblings of CCP wanting to control export of top tier frontier models.

> Of course, it still remains to be seen if they will keep publishing the weights at that point.

I mean if they don't then the US companies could just release their weights and chop the Chinese off at the knee's at that point (reversing what is currently happening).

Opening the weights isn't just a technical thing, it's a geopolitical decision which likely in part comes from the top of the CCCP (by either request or just implicitly allowing it).

Open weights is not a CCP thing. It's a business decision because Chinese labs do not have a good distribution strategy in the west/outside of China.

This is nonsense, at a policy level the Chinese government have actually been moving towards restricting larger models. I'd eat my hat if Europeans in particular aren't paying licenses for weights in a few years if it means getting to host the models on home soil, there is draft legislation (CADA) making its way through at the moment that would even require this in many circumstances.

> But the goal posts will move

On both frontiers and open. And lately the gap has been closing so if anything yeeting Anthropic etc is getting more viable over time

> will find new areas that have a need for absolute frontier

And they will keep it to themselves to extract the value for as long as possible.

Thank you! I only opened the comments to find this link.

We did the same thing, on much less powerful hardware but it has served us very well and we don't need to worry at all about the question whether or not the suppliers of inference services can be trusted long term with our data. The answer to that question in my opinion is 'absolutely not'.

Can share some details about your setup? Would probably be interesting for a lot of people.

Sure:

https://jacquesmattheij.com/local-inference-with-large-langu...

If you have more questions feel free to mail or ask here in the thread and I'll do my best to answer.

[delayed]

Thank you, I'll fix it right away.

That was a good read!

That was a lot of work to put it all together.

It's basically the document I wished I had when I started all this.

How do security conscious self-hosters sandbox their runtimes?

Your inference server is just an inference server that spits out tokens. You can use the same harness, etc. you’re basically just changing the API base URL.

By pulling the plug. This box simply has no pathway to the internet. I do pretend I have one and it's interesting to see what it tries to do. For instance, Qwen tends to try to hit ali related URLs for websearch and is very disappointed when the mothership is not responding. I've yet to catch a model on an exfiltration attempt, either they are so good at this that it escapes my scrutiny (possible but unlikely) or they are simply not doing it. I would bet on the last one. Just don't say you're working for the Tibet Liberation Frontier or the Keep Taiwan Independent movement.

Why would you sandbox the runtime? The harness is where things go awry when it does, when the agents have access to more than they should and start deleting stuff. People aren't expecting the weights themselves to be "infected by malware" or whatever.

This is the right call - there's little proof that the quality gap between open models and OAI/Anthropic stuff translates to business results, but you surrender all control of your data to a third party.

The last cause might be a straight up blocker - certain financial regulations straight up prevent you from giving third parties to your data (esp outside national/EU jurisdiction), no matter what documents you sign.

This is probably only relevant if you go over the 150 engineers putting you in the enterprise bucket and forcing you to pay by the token rather than the subsidized Claude code plans. If you need 150 engineers all with frontier model accounts maybe you are doing something wrong ;)

If you want zero data retention you have to go api only.

I feel ZDR is the same as the EU data privacy shield

> If you want zero data retention you have to go api only.

Time for the daily reminder that ZDR is a pinky-promise.

ZDR is a contract clause. That means:

    - It is subject to the rest of the contract, e.g. the good old "subject to lawful request or court order" carve-out clauses
    - It is subject to random provider exceptions, e.g. see Anthropic and Fable.
    - You have no way of verifying it beyond "trust me bro".

Basically you the only options are:

     - Run your own hardware which is obviously, the ideal situation
     - If you can't run your own hardware, use confidential computing which runs the AI in secure enclave, e.g. Privatemode[1] (Germany) or Tinfoil[2] (US)

[1] https://www.privatemode.ai/ [2] https://tinfoil.sh/

For my companies purposes US contract law is risk mitigating enough.

> For my companies purposes US contract law is risk mitigating enough.

Where I work, US exposure is a risk in itself. CLOUD act, PATRIOT act and then the general political situation in the US.

We have been working hard to minimize US exposure to the maximum extent possible because the risk has gone from theoretical to actual.

Which is why everybody buys tokens from Amazon or Microsoft or Google, not anthropic or OpenAI.

Because you already have your BI and mails on these providers.

> Which is why everybody buys tokens from Amazon or Microsoft or Google, not anthropic or OpenAI.

Erm. Exactly the same thing applies ? AWS, Azure or "direct" ... its all still "trust me dude".

> Because you already have your BI and mails on these providers.

You might. Lots of people don't. My $work and many others included.

How do you verify that privatemode.ai or tinfoil.sh are actually running the AI in a secure enclave? If you're the untrusting type, there's no way for them to prove they're not faking it being run on a fake secure enclave. If you're the really untrusting trusting type, "On Trusting Trust" should be your guidebook and local hardware won't save you either.

For the past several months about 90% of the work we do goes to open source models. If you ignore the noise around 'cheap' etc, you might find that with a little extra effort to more precisely define what you need not only do these models generally work as well, and they will generally more closely follow instructions - on longer timeframes this is actually pretty helpful.

They are less good on ambiguous prompts like 'fix all the problems' (which to my chagrin is a very common prompt it turns out..)

Yes, that's a good point: you really need to work on the prompts, even if the harness is the same. Ambiguity is a real problem as is what we call 'the negative space', what isn't in the prompt but what is implied by the prompt to be undesirable. You need to spell that out in more detail it is not enough just to say 'do x' but you also need to make sure to say 'and not y or z', and knowing what y or z are requires a bit more insight into what hobby horses are encoded in a model.

To give one example: I'm doing a lot of stuff that involves hardware. When settings things up for fixing a bug I have to make sure to include that the hardware has been tested and works well. If I don't do that the model will almost certainly spend a ridiculous amount of time speculating about all the different ways in which the hardware could be broken (including theorizing about chip mask errors and all kinds of other obscure and rarely seen in practice defects). It's going to end up telling itself that for sure the 500 lines it just produced are absolutely perfect and the device that is on the i2c bus is broken in a way that it just really can't progress with solving the issue because there is nothing wrong with the code.

These kinds of blind-spots are funny when they happen once in a month but when they happen regularly it begins to be irritating. Adding a line or two to the prompt will take care of the issue. Frontier models tend to be better at this sort of thing.

Us too! Hardware is very interesting - I find in fact that 'smart' (aka expensive) models can overthink complex hardware state machines and end up doing atrocious code because it will try to handle every possible edge case (unbidden, mind you). What has worked very well is harvesting and aggregating lessons and feeding this automatically into our pi sessions, this allows the models which listen to instructions to avoid a lot of time burning tokens on internal dialogue to ascertain the nature of reality.. but the expensive ones: a) they dont listen b) they ironically know too much!

Funny, I wonder how many people are retracing the same pathways. Another interesting one: asking for reboots or resets to start from a 'known state' even though this will not change a thing. But it's a quick path out of the trouble spot that might just work so why not try it. I've added the following line: "The hardware is on a spaceship and is impossible to access by an operator and for some reason they forgot to include reset logic, so we have to work with what we've got.". I've been gaslit by AI so often I figure I can return the favor without penalty.

IDK why/how this works but by giving it a standardised set of apis to interface with the hardware (read logs, restart it and so on) we find models tend to behave much better ... it wont randomly do stuff it feels is appropriate for a debugging session.

At this point though our context is so dynamic because of lesson propogation its hard to think about a model on its own - a vanilla harness (which we were forced to use recently because the memories system went down for a bit) just felt so different!

I have the agent software set up to log every byte in and out of the model into a file, time stamped and which side of the conversation was active as well as all of the tool calls, parameters and stdin/out of the tools. Every week or so I start a job to process that log file and to extract useful lessons for future sessions. These go into a LESSONS.md file that gets read by every agent session on start-up.

thats how I started, but was harder with multiple people - we opensourced our proxy: https://piproxy.latchlabs.dev/

We have a tiny team, so we can get away with simpler solutions, but yours looks pretty good. I may have to try it :)

That is very interesting. So far I am extremely careful to avoid putting negative statements into my instructions, to avoid poisoning the AI's thinking with exactly the concepts I do not want it to have. This might be from my use of much older models where if you say "do not X", then "X" gets put into the context and the "do not" kind of gets lost. It was especially visible with stable diffusion where you would ask for a picture without a feature and get exactly that feature. Maybe current models are better at that.

DS 4.1 flash 1017 is June's sota at such low pricing it doesn't matter how many tokens you're using.

"the gains showed up when tokens stopped being rationed" - I can totally see that. I am leading (as an external consultant) a larger dev team for a huge insurance company. The devs are quite junior or in general more the "I became a sw-eng here because that company is the only big employer in a 1h radius" type of people (meaning: No rockstar developers). Most of them do not play with AI in their spare time, and they have been told to spend claude tokens visely. Everybody is too afraid now to play with it and explore during work hours for fear of being token shamed.

At the same time, most of their junior devwork can be replaced completely by AI - so they really should start to learn how to work with it. I recommended everybody to get a private 20€/month claude subscription and use it for toy projects. I have younger devs (early 20s) in the team that told me they won't spend private money, or simply have no usecase in their private life for AI. We talk about software engineers here, and IMHO most of them are in denial of the fact that their skills have mostly become worthless if they don't start to pivot to something valueable.

If you were to take away my work's $500 budget for Opus 5.5 and replace it with unlimited use of GLM-5.3, it would be a huge downgrade in my opinion.

If your employer is so cheap as to spend a $100 or less per month per engineer, while being large enough not to have access to the subscription plans, a move to self hosted or cheap providers does make sense in my view.

The thing is the have been told to use the tokens wisely, but no limit is communicated. Still everybody is hesitant. To be fair, the company runs claude on AWS bedrock for data privacy requirements at a different pricelevel. But I guess that is the drawback for german job-security: each and all of them know they basically can't get laid off.

I also attribute a lot of it to be just plain outright denial. I have a feeling a lot of people don't WANT to understand that their coding skills they acquired over years have dropped a lot in worth. It is the "as long as I don't step on a scale, I can't be overweight" kind of denial. You are afraid to see what is behind the curtain, so you simply don't do it.

> But I guess that is the drawback for german job-security: each and all of them know they basically can't get laid off.

Help me understand what exactly is the drawback you’re mentioning. In your previous comment you were suggesting these people spend their own personal money and personal time on projects to benefit their work time. They (quite reasonably, it seems from the limited context) decided that no, they won’t dedicate life and resources from outside work to more work. Fortunately, it seems they have some job security (German, but probably some from the EU as well) which allows them to make whatever choice they believe is healthier to them, without fearing repercussions. What’s the drawback?

No you misunderstood my comment. I suggested they spent their private money - 20€ a month for the lowest Max flatrate tier - because to me it is obvious that you absolutely need to develop the skill to work with AI. They only have the "pay-per-token" model on their work accounts and laptops, because that is what large corporations get - no flatrate. That leads them to not work with AI. Most of them just refuse to get that 20€/month subscription to experiment with it on their private laptops. I didn't say private time - company policy is no US cloud services for work related topics so they would have to install it on their private computers. They totally could work on toy projects during their time, as the team is in homeoffice a couple a days per week.

Thank you for attempting to clarify, but it doesn’t seem to me like I misunderstood the first time. Seems you are indeed suggesting they spend their own personal money for a skill to be used on work time, and that they use their work time for personal projects to ultimately benefit that work time.

I don’t see how working for a public entity is relevant. If anything, I sure hope (but I’m not deluding myself) the people who work for public entities work slowly, deliberately, and methodically. The last thing I want from public entities is a focus on “velocity” and merging code the authors don’t understand.

> and that they use their work time for personal projects to ultimately benefit that work time.

That is exactly what I am saying and I would advise them. If you are early 20s right now and a software dev, just waiting for your employer (a large, slow enterprise) to train you an AI seems like a career killer. The whole attitude of "I landed a job with job security, so I am not doing a thing more than I have to" is what drives germany into the ground...

> "I landed a job with job security, so I am not doing a thing more than I have to" is what drives germany into the ground...

That seems strange to me. A country is made of people, and people are more than their jobs. The fixation on work above everything to benefit an entity whose only purpose is to make money and that would chew and spit you out at the first opportunity seems unhealthy to me. I’d much rather we worked on making the people happy and not concentrate so much wealth in the hands of a few, that seems to me like a better way to improve a country. But hey, happy to disagree. Thank you for engaging and expressing and clarifying your views.

But why should pay everyone else for a lazy/inefficient person? Because that what the salary a person becomes is: someone elses money.

I used an immense amount of personal time and money to make me a better employee, it was called getting a degree from a university.

I still sometimes spend a bit of personal money and time to tinker with technology, sometimes on the basis that it'll probably be good to know some of this stuff. It's like being surprised a chef likes to experiment a bit in their kitchen at home.

Their point is with total job security there's no real drive to actually be good at your job and little incentive to improve. Just be good enough to not be labeled entirely incompetent and truly unfit for the job, not any better. Would you go to a doctor who hasn't bothered learning anything new about medicine for decades and still practices like it's the 1950s, or one which continually evolved their practice to stick with modern times?

> I used an immense amount of personal time and money to make me a better employee, it was called getting a degree from a university.

You went to university to become a better employee?! Please don’t take this the wrong way, but I find that notion profoundly sad. Most people I interacted with in college did it for the experience, and I’m happy to report many are better people for it.

But hey, I’m not going to criticise you for doing that, you made your choice about what was right for you and that’s all I’m saying, that it is a good thing that people Have that choice. It’s good that you can do that if you want but also not do that if you don’t want.

> I still sometimes spend a bit of personal money and time to tinker with technology

So do I. And it’s always best when I can decide to do it or not do it, depending on the rest of the context in my life.

> It's like being surprised a chef likes to experiment a bit in their kitchen at home.

No, no, it’s nothing like that. Because I’m not at all surprised that people do it (like I said, I do it myself), all I’m saying is that it seems positive to me that you have the choice.

> Their point is with total job security there's no real drive to actually be good at your job and little incentive to improve.

Aren’t you yourself proof that that is not true? And nothing indicates those people aren’t getting better, they’re just not doing the specific thing the other person suggested as a means to that end. That’s not at all the same thing.

> Would you go to a doctor who hasn't bothered learning anything new about medicine for decades and still practices like it's the 1950s, or one which continually evolved their practice to stick with modern times?

Ooh, boy. If you think most doctors out there are continually improving, I have some very bad news for you. Those who do are the exception, not the norm.

> You went to university to become a better employee?!

Not the sole reason, no. But it's not like that wasn't a reason, it was certainly a big one. And that's an incredibly common reason to go. And no, I didn't even get a degree in the field I'm working in currently, but the things I learned getting the degree have definitely helped me be far more effective at work.

> Aren’t you yourself proof that that is not true?

No, because I'm not a government employee with near total job certainty. If I'm not keeping up, I'm falling behind, and will likely find myself replaced in short order. There are millions of people in the world who can do what I do, and what I do can be done anywhere there's a data signal.

> nothing indicates those people aren’t getting better, they’re just not doing the specific thing the other person suggested as a means to that end. That’s not at all the same thing.

Sure. Believe what you will. I've worked with countless people who have this mentality. Giving these people more job security guarantees they'll care even less about their job performance and growth. Having the job is 100% about getting a check on the pay period, not about actually doing anything useful.

If someone doesn't think €20 and a few hours spread over a month isn't worth it to be better at their job, I can't imagine they're really working hard at improving any other ways.

> If you think most doctors out there are continually improving

I don't think that. I agree many stagnate. And I usually choose to avoid those which seem to be stagnating.

It’s honestly a bit hard to continue the conversation with you, because it already seemed to me like some of your arguments where bordering on bad faith in the previous comment (but I gave you the benefit of the doubt) and I get that impression in this comment as well. But I don’t want to have an internet argument, I don’t care about “winning”, I want to have a conversation. I sometimes fail at that and I don’t like when that happens, so I’ll try to cover broad strokes instead of quoting specific sections.

You seem to be suggesting that being a government employee gives you near total job certainty and that to have that job security you need to be a government employee. That is not the case. At all. Maybe where you live, but certainly not everywhere. And yes, I am keenly aware of highly incompetent people (in healthcare no less!) who have job security. Is that a good situation? No. But throwing the baby out with the bathwater and making it a free-for-all where everyone is stressed all the time and can be kicked to the curb on a whim isn’t a better way to properly advance a society. Not if you care about something other than the unfettered pursuit of profit, anyway, and I believe we should care about more than that.

Because you edited your previous post, I’ll point out that I find the new examples (teachers and cops) to be even worse. Policemen are notorious for abuses of powers with no repercussions, I honestly don’t know what to tell you if you’re not aware of that. Teachers get the short end of the stick, and the best ones are seldom recognised. Again, if you're not aware of that I’m not really sure what to tell you.

Also, of course it’s not good to have incompetent doctors, and of course that was not the argument. I invite you to steel man if you want to have a productive conversation. Nowhere am I saying unrestricted job security is the right solution, I am instead pointing out that giving all the power to the employer causes worse problems, and that in general we should care about people’s well-being instead of trying to “advance” and being afraid of “being left behind” all the time. That’s not healthy, and it is stupid to chase “progress” if not to improve lives.

> The whole attitude of "I landed a job with job security, so I am not doing a thing more than I have to" is what drives germany into the ground

I'm not in Germany, but I definitely see this attitude in lots of positions in what I see as jobs with excessive job security.

> Policemen are notorious for abuses of powers with no repercussions

So wouldn't it be good if they potentially had less job security so they'd actually face some real repercussions for their bad actions? Instead they often have some of the strongest job protections out there. They can sometimes even murder people and still collect a full paycheck while cooling off for a few weeks.

> Teachers get the short end of the stick, and the best ones are seldom recognised

I agree. And once again, lots of job security programs ensure the ones continuing to scrape by get promoted and kept while those actually trying just face burn out. I definitely think teachers in general should be paid more and should have good incentive systems (of which there are lots of bad incentive systems, it's a tricky thing to do well in this space).

> in general we should care about people’s well-being instead of trying to “advance” and being afraid of “being left behind” all the time.

I truly do agree we should try and have a good balance of work/life, and I do agree tons of people in places like the US are way too overly focused on the work side of that. I'm just pointing out when you focus too much on the other side, all your industries are no longer competitive on the global marketplace and you'll find yourself a lot worse off. Having people who aren't effectively incentivized to improve (for whatever improve means in context) is not a good thing, I think we can both agree on that.

FWIW, that kind of near total job security leading to not caring about improving isn't inherently a government job kind of thing. I've known lots of people in private industry be effectively the same. The real connection is places with extreme bureaucracy. And I'm definitely not trying to bash all government employees, many truly are hard workers who care.

What exactly is the skill you acquire by using more AI? If the models get smarter, isnt it a waste of time to get good at the current gen?

The skill of getting familiar with AI use, how to use it, when to use, learning when NOT to use it, is the skill that will land you your next job. Or would you in 2030 expect to hire software devs that never touched AI?

I'm probably a worse programmer than I was before I started using frontier AI models, in the sense that I probably have a harder time reading, writing, and debugging code. And I doubt that my knowledge of the AI models themselves will have lasting practical value. But I know substantially more about UEFI, PCIDs, seL4, WebAssembly, GLSL, Service Workers, CORS, server-sent events, CSS, content-security policies, memory-hard PBKDFs, Android, numerical integration, Verilog, V8 security, Linux process namespaces, hypervisors with VT- x, multitouch user interface design, radio communications theory, power electronics, physical-layer communications protocols, NTSC video signals, TLA+, bounded model checking, generative testing, theorem provers, FPGAs, Linux system call overhead, Unix mailbox formats, physical modeling synthesis, damped resonance, KaTeX, geology, dishwashing detergent, microbiology, Cyclone, thermodynamics, X-Windows, localStorage, AVX, and IndexedDB, for example.

I think this comment does a good job of explaining how the breadth of what software engineers do is going to change even more. And again, we are still early. Find a domain to work in!

I don't know, this might be year #81 of the 82 years in which humans create software. Or it might be year #81 of 82 billion years. Anybody who tells you they know what's going to happen is full of shit.

I do know it's year 1 of n of me being crazy productive. I was already a good developer. Turns out I'm a damn good business analyst/bot farmer/and developer.

[delayed]

That's the problem then - if the actual amount the company wants each employee to spend is maximum €200/mo, $500/mo or whatever then the company should set that as a limit and people will use it.

Saying "you can use as much as you want but don't use too much or you'll get in trouble" is obviously confusing and is going to result in most people barely using it

> If you were to take away my work's $500 budget for Opus 5.5 and replace it with unlimited use of GLM-5.3, it would be a huge downgrade in my opinion.

I am perfectly able to do the tasks with own custom harness and unlimited local Qwen 3.8 27B with 150t/s. Harness needs search capability and more fine-tuning for the job.

You need less words to describe the intent with Astra/Opus 5.5, and it automatically usually observes the best practice that you don't need to explicitly tell, but with only a slight more effort you can get the job done. I have even made security audits with Qwen 3.8 27B against projects fully written with Claude Opus and it is a bit mind-boggling, but there have been tens of vulnerabilities, identified by this model alone. This excludes the issues found when guiding more myself.

I’ve been going back and forth between opus and local Qwen 27B and a good number of times opus makes absolutely stupid things compared to Qwen.

For a large number of tasks you definitely don’t need a frontier model but a good harness around it.

If you don’t can’t/want to experiment with local model, have Claude implement the exact same feature with haiku/sonnet/opus and see what the differences are. You might be surprised.

I've used extensively everything between Opus 4.6, GPT-5.5, Fable 5, Opus 5, GPT-5.6 Sol, GPT-6 Astra, 6.1 Sol and Opus 5.5.

To a lesser degree also Sonnet and Luna.

And when I say extensively I mean 500h+ on a side project in addition to my full-time job.

Even the difference between 5.6 Sol, which is miles ahead of local Qwen 3.8 27B, and Opus 5.5 is immediately obvious in the quality of the results.

Having been stuck on 5.6 Sol at my job while Opus 5.5 was available at home, I can say with full confidence that the difference is significant, even if I weren't looking at the output at all.

The way it understands my requests, forms a hypothesis, verifies the result and finishes the change is night and day different. This happens while 5.6 Sol is still flailing around somewhere in the monorepo struggling with even locating the relevant code.

I use a mix of GPT/Codex, Claude, and DeepSeek for my workload (in that order of volume).

Each has their own characteristics and honestly you just get used to prompting each a bit differently.

Even better is to use an orchestrator (I switched to Paseo a few weeks back) with an agent "team" (in Paseo, these are just profiles that the orchestrator initiates). Then the orchestrator writes super detailed prompts to each sub-agent on your behalf and I find I get better results this way with less effort. I would describe it as "more autonomy".

Are you not worried about this being a sort of one-way street? Do you have any strategies to avoid ending up with a codebase you cannot work on without "AI"? when I think insurance company, I would assume the value seems to be more in the trust you build up over decades of with customers, not producing lots of code faster.

Well I am still in the loop as a senior dev and I get to review everyones code before it gets merged. Official guidelines is that everybody is responsible for the code they produce (with or without AI) and must understand it when they give it to the reviewer. I've been doiing this 3+ years at this company, so before the whole AI thing landed. I am explicitly not allowed to implement features myself (because the idea is the juniors learn and get trained).

Now, within the last 6-12 months of AI advancement, I think you could lay off half or even 3/4 of the team if I was explicitly allowed to implement features myself using AI. But that is not the current policy.

As for worries "ending up on a codebase you can't manage without AI" - my personal take is: No, I am not worried. Because going forward AI will always be available in the development workflow. That is my own opinion, not the corporation standpoint, however.

Being unable to work on a codebase without AI is not necessary. Working with models you can fit on consumer hardware is already significantly more productive than rawdogging code. It's similar to "you won't always have a calculator with you" from math teachers, see how that turned out.

A way bigger problem is a codebase that you can't effectively work on with AI, which is a lot of legacy humanslop. And things that make a codebase more suited for AI, like documentation that provides relevant domain knowledge and context for requirements, automated quality control, and test coverage, also make a codebase easier to work on for a human too.

You get the same result when you calculate something by hand or with a calculator.

And that whole "legacy humanslop" paragraph is just stating the obvious with an air of spite towards humans.

People who can't bother to reason get angry when you say "hey, why put all eggs into one basket", episode N. Boring.

> ... and IMHO most of them are in denial of the fact that their skills have mostly become worthless if they don't start to pivot to something valueable.

But your comment says it several time: the valuable skill is being already a seasoned dev.

Then AI is superpower.

It's for the junior that things sucks big times.

I see it all around me: senior devs are empowered while juniors are not even hired anymore.

It is a shock going through the industry, and we are just starting. I think future software engineers will need more training years and more unpaid time getting skilled. More like surgeons, who need almost 10 years of training until we actually let them operate on live patients.

Given that we are seeing a big retirement wave of the boomers, a lot of seasoned devs will retire in the next 10 years. So there should be enough demand, and a big enough paycheck for that decade long training to make its moneys worth.

[delayed]

Why doesn't your company just get each of them a 20€ subscription instead of paying per token?

For compliance reasons, companies pay for the enterprise account, and the enterprise account uses API pricing.

Well that is a great idea, and if I were management, I would give everyone a 240€ claude-giftcard-subscription that they can use privately on their private computers. There is however the danger that they would start using their private accounts to fill with company data (which potentially is against GDPR regulations - hence the amazon bedrock deployment).

> most of their junior devwork can be replaced completely by AI - so they really should start to learn how to work with it. I recommended everybody to get a private 20€/month claude subscription

I'm struggling to wrap my head around this. "This technology will completely replace you and lead to your unemployment, so you should fund it out of your own pocket"? Is that what you're saying? If so, can you please explain your reasoning?

AI will not replace everyone. There will be people steering the AI, and using it. If you are not one of those people, you WILL be laid off. It is like office workers in the 90s thinking they don't need to learn computers and hold on to their typewriters.

Hey Opus 9, you are an insurance carrier offering web-based policies and servuces across the EU. Available tools: german_notary_api, bafin_compliance.

Here is €1B in initial capital. Go!

Funny, but insurances are a good example. Most of what insurances do is administration. Like almost everyone's job there is to sit in front of a computer all day. Lets see how much of that will still be done by humans in 10 years. I think it is one of the field most impacted by layoffs, as they manufacture nothing.

This is a sane version of what “sovereign” AI can look like for many. In Canada we’ve just had this national plan to overpay for an inferior AI product in the name of sovereignty. Even if GLM is slightly behind, this is likely close to the Pareto frontier.

They also could consider putting this on openrouter when they have spare capacity to make the most effective use of the hardware.

GLM 5.3 is far ahead of anything Cohere has ever released, and it's relatively easy to host.

Anthropic's missing last quarter results in a nutshell.

At dayjob we're also using self-hosted GLM-5.3 via our flavour of pi.dev but teams still have discretionary power over using Claude Code or not (there's a team's budget to spend on whatever the team decides). Some teams like mine still like to use Claude Code for a lot of the work but most stuff doesn't require the latest and greatest LLM and can be done with "lesser" LLMS by simply throwing a lot of tokens and compute to it.

My mid size German client (100M revenue) is also moving lots of work to on premise LLMs.

I have no clue how do the lunatics think there's any moat in models.

Here's the part that matters:

What does "blended $0.30 per 1M tokens" mean - is that input or output or average across the two? (asking as am trying to benchmark that cost against public GLM providers on Openrouter)

It's an average across the two.

One would still have to guess what the ratio of input tokens to output tokens is. (I would say 10:1 off the cuff, assuming this is editing an existing codebase)

> the gains showed up when tokens stopped being rationed

My whole company (Amazon) has unlimited tokens, and the only thing it has led to is insane amounts of slop. Illusion of productivity.

If I wanted to buy 56 B200's and had $3 million, when could I get delivery?

I would sacrifice a few benchmark points for higher tok/s actually, especially if more bandwidth as well so I can scale up swarms, then I can likely compensate for any drop and get more mileage out of it.