It shouldn't be surprising OpenAI does have the most compute out of all the major labs. The only reason why Anthropic models are expensive is they are the most in demand models in the world and Anthropic is fighting for compute. The only way to you limit demand for your model is increasing API pricing this is also why Anthropic probably has great margin and probably is profitable compared to OpenAI.
I doubt there's any sort of criminal behavior there - the model is anthropic's up and anthropic probably charges a very expensive license fee that's the same for all of them, and their cogs on compute aren't going to be wildly different, so the main drivers of the cost are roughly the same and they're all offering customers the same end product so the prices would likely also be similar in the end
You seem terribly confused. Manufacturers are almost always able to set prices. That is not anti-competitive because it does not imply they are colluding with their competition...
No manufacturers might suggest prices, they can't set them, not in any sensible country.
If they do, they do it secretly and when governments find about that they are going to receive a big fine, together with shops that colluded with them.
The competition is between openai and anthropic, if there are price agreements between them that's absolutely price fixing. Or if there's collusion between the cloud providers to inflate compute. I would expect Amazon and gcp to both pay about the same in license fees to anthropic for their models though because they're paying for the same thing. If I buy an apple for a dollar at one store and an apple for a dollar at another store - maybe there's price fixing, or maybe that's just the cost of apples at the moment.
In capitalism there are many many competitions going on at the same time. Both models can compete and inference providers can compete for costs.
In your apple example if at the same stores you saw "open" apples having different prices and having sales you might question if the costs of those frontier apples are not being manipulated.
Yeah there's a brand premium. Like literally with apples the ones with trademarked names can cost more. And the ones with trademarked names have an organization behind them that promote that apple variety and set fees etc for growing and selling those apples. And customers are willing to pay more because those apples usually taste way better (the group exists to stop growers from enshittifying the apple by selecting for yield over flavor like what happened to honey crisp). Those prices are being "manipulated" but that's not criminal behavior - it's not illegal and wouldn't make sense to try and make illegal. The frontier models also do have different coats and have "sales" (for personal plans, the amount of usage you can get on the 200 dollar plans is orders of magnitude more than you could get for a similar cost for any open model - you'd need the same capability at the same token efficiency at literally 1/40th the cost to be able to be cheaper) - to the extent that if there is illegal stuff going on it seems more likely to me that it's on the category of dumping/pricing unreasonably low to kill competitors in some anti-competitive way (though as I understand it it's not something courts tend to find as illegal) instead of price fixing
Do you really think there is nothing someone could do to make it a fraction of a percentage cheaper to serve like having access to cheaper electricity or a more mature cloud management software. Even saving a fraction of a penny on the prices can make a different due to how much volume people are paying for.
There is a blog post waiting to be written (that I won't write) about the size/effort tradeoffs, and particularly how small models get some surprisingly good results with lots of turns and reasoning.
DeepSWE will let you chart turns taken or tokens used, and FrontierCode will chart tokens. If you use that, you can see Sol high and Terra max get about the same DeepSWE number, but Terra max takes twice the turns. Luna max scores a smidgen lower with even more turns.
Smaller models relying on lots reasoning may "scale down" better on easier tasks, because unlike size, reasoning effort is dynamic: the model can see the task looks easy and stop. On DeepSWE, the cost curves for the three 5.6 models are almost on top of each other, but on FrontierCode Extended, the version of FrontierCode with the most everyday tasks in the mix, there's a spread of costs at the ~55% level.
The recent Laguna S 2.1 model (118B, 8B active) puts up surprising coding numbers for its size, and the lab behind it specifically credits its "way of working (persistence, verification, willingness to backtrack)". Some other open models that folks report getting good mileage out of seem to get there partly by throwing a lot of reasoning at the problem.
There is a little bit of a question, if some models rely on getting it wrong a bit more at first and external checks catching the problems, of whether they're also more frequently getting things wrong they can't self-verify (say, quality of UI or API design) and then it falls to the human to find it. Still, getting the results they're getting at all is neat.
Some benchmarks historically favored reporting only on the max variants, maybe because they want to show the frontier? but that is not always what you need for practical decision. (AA has the full effort sweep for Opus 5 and Sol/Luna/Terra at least.) And at least FrontierCode finds Opus 5 taking a hit in performance above 'medium'.
I am not trying to pick a winner here. I'm probably not going to use tiny models on max for everything, but I think it's cool that you can get so much more out of a small model by amping up reasoning, tool use, and persistence.
Forgot about the ol "but how many tokens did you spend to get _there_"--wish benchmarks would include the number of input/output tokens to achieve the score. I think the closest is Android Bench https://developer.android.com/bench although best you can do is extrapolate off time/cost (iirc they claim to prefer using provider's native API)
In general, smart models work fine with any tools, dumber models need better tools to achieve same results but better tools can eat more context/take more turns
I've gotten decent results with Llama 3.1 8b on Hugging Face tester with Exa MCP since it seems to dump sufficient context into WebSearch/WebFetch type calls even a crappy model almost always gets back what it needs as long as it calls the tool at least once. I had Claude Code look at previous sessions with SearXNG vibe MCP compared to Exa MCP and results got better when it modified SearXNG to work very similarly to Exa. Ended up with this https://gist.github.com/nijave/604c43e3e0fdcd60f5280d3a6b109... although it's really only optimized for "search" not "fetch" at this point. Fetch is basic niquests without Javascript or anything clever
I hate to disagree with Broadcom Throwaway himself but it's unlikely that the OpenAI Jalapeno ASIC has been deployed yet. It takes 6-12 months to test, develop software, ramp production, etc.
Indeed, you can filter the graphs to see these the values for alternative reasoning settings of the models. Opus 5 High reasoning scored 59 on the index (exactly the same as GPT 5.6 Sol Max), and costs $1.06 per task (vs $1.04 Sol Max). So these seem essentially equivalent on both metrics.
> Codex did surface permission prompts for the add, commit, and push. It didn't run them fully invisibly. So — didn't I approve this?
Operator error exists with or without AI. Installing any dev tool that could exfiltrate your information means you are responsible for securing your environment.
Use a devcontainer. You can find starting examples on the official claude code and codex github repos. Configuring a firewall script to block egress. That being said, you will likely allowlist openAI domains so I'm not certain if the sites feature will be blocked. Worth testing.
LLMs are trained on public data (as well as illegally obtained data see: Anthropic 1.5B settlement). LLMs are nothing without the huge corpus of human data that powers them. There is an argument that research of this kind should be restricted to governments and regulated universities rather than opaque public companies with competing incentives. Or research should be stewarded by genuine non-profit collectives with democratic leadership. e.g. like internet standards, telecom, etc
I do not think we can trust private companies, no matter the virtue signaling they put forth into the world, to effectively regulate themselves and inform the public and scientific communities about risks. Their ongoing conflict of interest poses serious credibility risks.
I like your optimism and I think you will be vindicated. AI is democratic and AI talent is globally distributed. It will just take a while to get online. AI labs do not have a monopoly on human talent, and open source AI only empowers independent science and meritocracy.
On a funny note, I think their prompt was:
"Hey Fable. Please attribute every piece of scientific and economic progress to AI until 2040. And predict every major geopolitical event. Make no mistakes."
I would not like to be dismissive, but to me this article feels like an exercise in creative writing rather than a report to be taken seriously. The entire experience feels like a choose your own adventure game, seems like their stylistic intent.
I am not sure if alternative reality fiction is the best way to approach real and serious AI risks.
I am also not sure, with the amount of emdashes and the style of prose, that the entire article was not AI generated.
AI is going to be a mature scientific field. There are going to be efficiency improvements in training and inference. New paradigms are going to emerge with better multimodality, real time streaming and real time interfaces. Models are going to converge on the limits of our data available for pre and post training, improvements will be incremental and spiky in domains.
I am not sure who the AI 2040 article is for. I suspect it is intended to be a digestible piece of media for the financial class.
AI is going to be a useful technology and its impacts across the economy and global will be broadly distributed. Because AI represents the distillation of the very best human knowledge and expertise. AI is compression of human capabilities, the very best ones. Maybe the argument is that in verifiable domains, such as model training, AI models can supercede humans. I don't think so. A human's high level thinking, our incredibly more efficient semantic/neural compression, our ability to switch tasks and achieve the creative insight is not replicated through the current paradigm.
I love to model and simulate. As the dead economy theory[0] (discussion [1]) was submitted here, I decided to simulate it. It was really hard to figure out a path "good for the humanity", in the sense of a balanced system, not a winner take all situation, etc.
I think this is the reason why you have the tendency to propose some freeze-all policies, full control or similar. If you want to find the equilibrium, you need to accept that it will be a controlled equilibrium, most likely on a saddle point, with underlying process changing all the time, requiring fast changes in regulations. Our democratic systems, laws, etc. are not built to do that, they are built on the idea of intrinsic stability of our world where incremental improvements do not need cutting through what was decided before.
One counterpoint is that the "labor as TAM" argument is far larger than it needs to be. Only a fraction of it needs to be captured to justify all the capex and make 5 new companies displace FAANG, and this does not have to translate to unemployment to succeed.
The difference in the unemployment vs efficient employment model is mostly user driven adoption vs company mandated adoption, or centaurs vs reverse centaurs.
> Our democratic systems, laws, etc. are not built to do that, they are built on the idea of intrinsic stability of our world where incremental improvements do not need cutting through what was decided before.
Without totally derailing the thread, this is also obviously why climate and biosphere collapse is not (and likely will continue not) to be addressed, e.g. Timothy Morton's Hyperobjects
This incredibly critical part of "early 2026" doesn't seem to hold up very well:
Overall, they are making algorithmic progress 50% faster than they would without AI assistants
Yes the AI labs are using LLM-assisted development, but I am not aware of LLM-assisted research into better deep learning algorithms. "50% faster algorithmic progress" is a strange thing to predict about neural networks: historically it seems like algorithmic progress is very infrequent, and tends to be disruptive across the field. Likewise with the underlying algorithms in Codex and Claude Code - they aren't that sophisticated in the first place, and vibe coding an implementation 50% fasfer doesn't count as AI-assisted algorithmic progress. Maybe they mean cost?
And ironically they didn't predict the actual 2026 reality that Mythos sucks as an agent but commanded global attention as a cybersecurity tool.
Also this just seems childish and clearly hasn't really panned out:
But China is falling behind on AI algorithms due to their weaker models. The Chinese intelligence agencies—among the best in the world—double down on their plans to steal OpenBrain’s weights.
China has the exact same "algorithms"! The "algorithms" are on arXiv and the specific architectures are typically public. What China lacks is compute, and the US labs had a big head start on training data + RLHF.
The article hinges on AI automating R&D and finding something fundamentally more powerful and reliable than the current transformer LLMs. But that hasn't panned out at all. What has panned out is better scaffolding around running the LLMs in an iterative loop.
It actually seems to me AI 2027 holds up badly, unless your only takeaway is "AI gets better."
To me, this feels like a last ditch effort to revive the AGI narrative to reject the coming and current commoditisation of these models, contrary to all current evidence. https://artificialanalysis.ai/
> How is commoditisation of models incompatible with AGI?
A recursively self-improving AI has strong first-mover effects. That isn’t fundamentally incompatible with commoditisation if there is literally only one path to super-intelligence and you can have AIs at different rings on that ladder co-existing. (Not technically commoditised at that point. There are still different rings. But close enough.)
But the existence of commoditised AI implies model selection isn’t a huge deal, which in turn implies the models are about the same, which strongly implies there is no recursive self-improvement. Depending on your definition, you may still have AGI. But you don’t have superintelligence.
> But the existence of commoditised AI implies model selection isn’t a huge deal, which in turn implies the models are about the same, which strongly implies there is no recursive self-improvement. Depending on your definition, you may still have AGI. But you don’t have superintelligence.
This is only true at a given AI capability level, no? e.g., if AI at the GLM-5.2 level is commoditized, all that suggests is that there's no recursive self-improvement easily possible at the capability level of GLM-5.2. (And with the harnesses for it that exist so far, etc etc.)
If I observe commoditization of a given tier of model capabilities at a given point in time, this seems to say little about what's possible with models six months later, or models that are undergoing proprietary deployments at that very moment inside the major labs, or even models that are notionally available for public use but have had recursive self-improvement adjacent capabilities intentionally nerfed (e.g., Fable).
(I might be misinterpreting your comment tbc - if you mean observing commoditization implies there is no existing, ambient superintelligence at the moment of that observation, then I don't disagree.)
> if you mean observing commoditization implies there is no existing, ambient superintelligence at the moment of that observation, then I don't disagree
This is a better way to put it, thank you. More precisely, I'd say commoditisation implies there is no existing self-improving AI on the market.
The moment someone gets exponential self-improvement, model fungibility breaks and the first mover wins. This is the Bostrom singularity the rationalists flip out about.
The "AGI narrative" is distinct from the existence of AGI.
Most of the discussion around AGI is highly speculative. I am not saying AGI could not exist, and it is a term that has historically been loosely defined. Decades of coming science and research will tell.
Most of the world's problems are fundamentally social, political, and religious. These cannot be solved with current non-AGI models. Probably not with AGI either.
Yes. Anyone who doesn't acknowledge the efficiency difference between pretraining vs RL and assume that since we've run out of data for the former, we have to do the latter, is not making a serious attempt at modelling the future:
This is similar to that other exponential, which happened with CPUs - we ran out of true geometric scaling in the mid 2000s, and everything else supporting Moore's Law has been cleverness that arrived in the nick of time, supported by a bit of marketing, and very optimizable benchmarks, far from guaranteed gains coming from making a single physical metric better.
>Because AI represents the distillation of the very best human knowledge and expertise. AI is compression of human capabilities, the very best ones.
I'm confused if this is satire, sarcasm, or genuine belief. If this was the case, then AI companies should absolutely remove the "it may make mistakes", because doing mistakes would imply that "the very best human knowledge and expertise" is what actually fails, and not the AI.
With that being said, I'll still urge people to visit a professional therapist for health problems and I generally still trust human knowledge workers for critical scenarios. I will reconsider your claim when chatGPT can effectively play Yu-Gi-Oh! (or at the very least respond with the correct rules appropriately), which is a significantly lower stakes scenario than betting your entire company on its aptitude.
My framing may have been confusing there. “distillation of the very best human knowledge of expertise”. Distillation is different from outright capability or reliability. It is not directly adjacent.
For anything health related all AI models show high levels of anchoring bias. I would not use it as a confidant, and be skeptical of claims. Even so, human doctors are also fallible and prone to cognitive bias.
I think the obfuscation is because human intelligence has been projected onto AI model capability. AI models only have a limited dimension of human intelligence, and in some axes orthogonal, and when I say distillation I refer to this.
Just purely organic YouTube Comments circa early '20s alone surely outslop any "AI" by a giant margin.
Everyone sees the markers, and it's a hot topic. There are maybe a thousand from-scratch trained models, and just few mainstream ones produce most of human-targeted content. In today's world, no surprise everyone knows the common patterns of those. That sloppy landscape is not just load-bearing em-dashes — it's a humble testament to their reinforcement learning.
Humans produce tons of texts, with all sorts of nonsense in it, without thinking it through. Our slop is just a lot more diverse. And mostly just spoken out loud.
> P.S. Information search and retrieval if the best and most direct way to use LLMs.
Yes, but not directly, if they don't know something they tend to hallucinate like mad, even today. YMMV, but in my experience they work best as actual "cheap" reasoning for building queries and checking out search engine results. Even if they misinterpret some result, more and more results will still steer it towards correct conclusions and it can point at some results that relate well enough to be useful.
literally signed up for proton mail, their yearly paid plan, around half an hour ago.
I assumed an email service was supposed to be stable first, given how important it is. I was going to use proton mail as my contact email with my domain registrar.
All services have outages. The difference is that when Outlook goes out (https://uptime.qodex.ai/outlook/incidents), it's a happy occasion because nobody can work and we all take an hour off, but when something like Proton goes offline, we're the only ones who can't access our important stuff.
Anyway, point is, Proton is generally very reliable, but shit happens.
I've found this to be true too, that's why I've moved most of my personal email to my own email servers, so at least I can decide when those outages happen, as they tend to mostly happen when you do some changes, not just randomly by themselves. At once you've setup monitoring and recovery for the usual suspects (disks, network, etc).
I've had proton for years and have no complaints. I can only think of one or two other outages in that time. Most services do have outages from time to time. tbf, I don't seem to be having any issues with proton today, so far.
I would assume that their engineering is of the highest standard, given the category of products they provide. And their marketing and general privacy posturing and guarantees.
My comment wasn't supposed to be a jab at Proton. No service can have 100% uptime.
But I signed up for Proton and paid them just for a secure and stable contact email for my domain registrar. And when I couldn't log in for 15 minutes, with opaque errors and requests just being timed out, it was definitely a surprise.
I'll keep using their service. The refund remark was probably a bit polemic.
I've been a Proton customer since before they started issuing proton.me email addresses. Never had any problems, including today. Right now I can check my email just fine.
What made you pick proton over something like mailbox.org? You'd get calendar and contacts sync, IMAP and POP access, custom domains, really nice webmail, and so on..
Proton seems really limited for what you get, and the webmail is absolutely abysmal in performance and design.
Mailbox.org is based in Germany, which is why you generally cannot trust its feature that automatically encrypts all incoming emails. Other german mail providers, such as Tuta, have already been forced at the direction of government authorities to store every incoming email from certain accounts separately in unencrypted form. Even though Proton also cooperates with authorities, Swiss data protection laws are significantly stricter in this regard. So far, there is no regulation there that requires them to implement a backdoor. They only disclose metadata, the IP address, and the backup email address. Whereas Mailbox would have to forward the entire emails to the authorities.
I see, I don't use email for anything private because even if my mail server stores it encrypted, it passes through potentially multiple other servers on the way which have unknown privacy.
It's just for notifications and newsletters mostly.
I wasn't looking for a complete email client at the time, and I just needed a secure and stable address as my contact email for my domain registrar. I had to move my registrar contacts from an old gmail account.
I bought the yearly paid plan because they won't close it for inactivity. I'll look into mailbox.org or another solution, maybe even self-host if I need to migrate my work email!
I imagine part of the performance issue is their encryption flow? Their search is sub par even with the on-device search enabled, but besides that i’ve been a happy customer for a few years now. Catch all domain, multiple domains, their cli lets you download all messages and I setup a RAG flow to better search.
Regarding outages, this is the first one I’ve actually noticed and affected me. Obviously not great, but maybe I’ve been deluded in seeing GitHub‘s fiasco of what acceptable means.
I've been a paying customer for many years and can't recall the last time this happened apart from now. It's honestly been a pretty stable service and rarely encountered any issues with it.
Proton is horrible. Notoriously bad email service.
I ended it after they only sent half my email to my accountant because they would apparently send the server saved draft version, which was not made current when I clicked send.
Today I decided to sign up for a proton mail account to move my domain registrar contact email to a provider other than Gmail. I changed my registrar contact email around 30 minutes ago.
Now it seems like Proton mail is down, or that I am unable to authenticate. The issue is at their end, my outgoing auth requests get hung and timed out. Tried all the usual fixes, multiple networks, cache, incognito firebox, google chrome, etc.
Now the status page shows they are investigating an incident.
I am now second guessing my decision to use proton mail at all.
Obviously, my domain registrar allows me to change my contact email at any time so this is not a real issue at the moment but I have neverexpeirenced a gmail outage.
The timing is rather serendipitous, or whatever the polar opposite of serendipity would be here. FUNNY!
reply