I tried the Fugu models with some real world tales in C# and unity using mcp and open code. I exhausted the $20 plan 5 hour window in one prompt to review my theme system and plan some color changes. So I upgraded to the $100 to see the implementation and result. Well the result was worse than Opus, incredibly slow, and I ended up exhausting the new 5 hour window and have used 35% of the weekly now and it hardly created something opus was able to do at a fraction of the time and cost.
Do what you wish with this info, but it seems to be a complete waste of $$.
Because Fugu is not an independent model. They just use multiple existing SaaS models from OpenAI, Anthropic etc in background, gather response and generate results based on these response.
They claims that combining the results of multiple AI models and generate final result by using their in-house proprietary model improve the quality than using the single backend model.
It cause all sorts of doubts like: is their in-house model really exists? Is their in-house model really capable?
Personally, even if their claim is correct, such feature can be easily implemented in client side like Claude Code etc, using the equally capable model from background model to generate the final result.
We provide a similar service for Godot instead of Unity, and 20$ plan being exhausted in one prompt on a top model like Opus sounds about right. That's the life when you pay API prices and can't afford 10x subsidies.
Is this true even for more targeted prompts that aren't about looking over an entire codebase or w/e? I just stick to my sub pricing and find good success on targeted requests, but I wonder if I would run up against things even then if not for subsidies.
I tested Fable through Cursor; asked for ideas on how to make a data website I have less "Claude-like" (IYKYK what are the usual tells), and it spun out the most useless, Claude-like CSS styling ever, wasting $40 in 10 minutes.
The website was created through Opus, so you could also say the results were worse than Opus. (This is just to say that I had the same experience using the US models, so perhaps those Asian models are Mythos-like lol)
I tested Fable for a whole day. And my experience was quite the opposite. I was blown away. Admittedly, I did not try it through a middleman like Cursor. I used Claude Code CLI.
Me too. It’s output was fabulous. And it acted like a senior engineer - actually coding up hypotheses, testing them, finding problems and presenting good, usable recommendations backed by solid evidence and wisdom. It can probably do most of my job, which gave me a bit of an existential crisis.
I’ve paused my Claude subscription until they bring it back. Opus makes mistakes constantly, on every level of abstraction. Every time I look closely at its work I find problems.
If you actually give it an example of the style it can copy it well. even just screenshots of other websites or UIs. It just sucks at producing it itself.
Well, in part because the phenomenon has been discussed on Web forums that (a) have at this point made their way back into training data and (b) are accessible in Web searches that the model can invoke. And in part because the model can "know" what its initial instinct is and "decide" to go against it.
I experienced the same exact thing, however, I will say that I had misconfigured it on `pi` at first.
I was using its chat endpoint not the responses with tool calling and everything, and I haven't tried it again since, learned that recently and I am planning on giving it another shot.
This is useful info. For the couple of days that Fable was live - it was clearly a step above Opus 4.8 and I was able to get 8-10 prompts in using my $20 plan.
Fugu Ultra [0] is not actually a model, it's a system (harness in the cloud?) that routes to several models, looks like it's a bit like OpenRouters Fusion [1].
"Rather than a single monolithic model, Fugu is a learned multi-agent orchestration system: a language model trained to route tasks across a swappable pool of underlying models and to recursively call instances of itself." - https://openrouter.ai/sakana/fugu-ultra
Asian ai startups have also no way to compare despite making bold claims, and one could argue the whole point of the trump intervention was to prohibit them from distilling faible.
I don't even look at benchmarks anymore. I just try different models as they're released on our large, proprietary, systems software codebases in real, shipping products or projects that will ship eventually. It's pretty clear which models help me do my job better or faster. I'm fortunate enough to have the token budget to use basically as much as I need, for now.
No need for benchmarks, evals, marketing, system cards or anything like that. I read the web for tips, practices and release announcements. My colleagues and I share our experiences with each other but beyond that, everything else is just noise.
I am not against AI but I do wonder how you guys handle the fact this leaks all your code and is stored forever on servers belonging to God knows who?
I “trust” OpenAI and Anthropic (somewhat) but to be honest I still feel only safe using it on code without any secret sauce whatsoever. Luckily that’s a lot of code, but still. I wonder how others are looking at this?
(FYI I feel the same about Github and we also don’t store our code there)
This is the way. Not that big of budget here. But if there’s something promising, I just try that for a month or so. But even then… at this moment I’m using z.ai models and those do the job. No need for anything else. So I’m staying until there is something new, same affordability, but a lot better. (Using a coding plan)
Well if they are hyped like Mythos then we can add that to the list of “like Mythos”. Perhaps what’s missing is their CEO warning the world that their model is too unsafe to be released on the internet and someone must stop them before it’s too late.
Has either of these companies released models before this? It's hard to believe that they could release a supposed Mythos-level model just out-of-the-blue. Deepseek, Z.ai, Alibaba/Qwen have been at this for a lot longer and have been releasing models with steadily increasing capabilities for about 18 months now. I find it hard to believe that these new companies would just suddenly release a Mythos-level model without releasing anything prior.
But did they deliver anything yet?
There are many startups with notable founders that are not being able to get models launched. And even if they are able to launch, so they get PMF.
In case of Sakana, they clearly focus on the Japanese market an have buildup a good pipeline on sovereign AI. But similar to Aleph Alpha or to a certain extend Mistral, I don't see how they can keep up.
I live in Japan and yet can't seem to pay for their API in JPY... I bet their enterprise customers don't have that problem but it was pretty annoying given "AI in Japan" appears to be there only selling point.
The only way they can compete in Japan is the enterprise-game. Their partnership with Daiwa and MUFG is probably exactly what they should be doing. I doubt that they get that far though. Like Mistral who has partnerships with BNP Paribas and Airbus, they deploy in an on-premise or private cloud and in those settings, their models make good PoC's but cant compete where it gets interesting -- Workspace Agents. If you look at them from a Chatbot over RAG perspective, maybe they can do it. But that's tech from 2023.
I think every country/major country at the government level will back any home grown talent in view of USA restrictions.
The models don't need to be as good, just good enough for the task. I am using a Claude Q3 2024 model mostly (Haiku 4.5) at present, and it delivers what I need.
How do we define "mythos level" exactly outside of marketing buzz? I don't even think the majority of us can access Mythos yet even to make a comparison.
Mythos is extreme hype. We are at a combo of authoritarian politicians peddling fear for power and tech bros trying to extract maximum investment returns.
We’re going to have many LLM/toolset combos that do what mythos does.
I'm expecting a ban of "foreign" llms due to "safety concerns" before the year is over.
It will have nothing to do with the actual performance. But anthropic has set the bar for mythos-like systems, and whatever meets that loosely defined bar will be unsafe for the public.
> Despite his research, Semmelweis's observations conflicted with the established scientific and medical opinions of the time and his ideas were rejected by the medical community. He could offer no theoretical explanation for his findings of reduced mortality due to hand-washing, and some doctors were offended at the suggestion that they should wash their hands and mocked him for it. In 1865, the increasingly outspoken Semmelweis allegedly suffered a nervous breakdown and was committed to an asylum by his colleagues. In the asylum, he was beaten by the guards. He died 14 days later from a gangrenous wound on his right hand that may have been caused by the beating.
It is materially better, but I didn’t feel a huge loss when it was yanked. My use case is large complex legacy modernization projects and its ability is definitely better than opus 4.8 at the job. But it’s more like an optimization, I could have a single or 2 pass in fable vs 8-10 with opus to arrive at the same solution.
First impression: Third-party benchmarks or gtfo. Personally, I've never heard of either of these companies before. We're just supposed to take their word that they've matched the best models on the market?
Sakana describes their model as a "Orchestration Model." Does that mean that it's actually a bunch of different models glued together?
Is it actually that hard to make good models or is it just about the amount of resources you have to do training? (This is an actual question, I really don't know.) I'm sure it's not trivial but does it really take world class secret knowledge to build off of the known existing techniques? I feel like there's tons of low hanging fruit still to explore, and time and resources are the limiting factor.
Like Zuckerberg, top talent may not work with a polarising character if they disagree with his behaviour. Space focused talent don't have many choices aside from SpaceX but ai companies are a plenty and a top AI person can pick and choose.
I suspect that Grok has been ironically lobotomized by pressures to correct its political views.
Similarly, I could imagine the Gemini folks working in a significantly more complex corporate climate, with different parts of Google pushing for different capability focuses. They are only lagging behind less than a year, so it isn't too large of a gap yet.
That said, the fact that Anthropic is currently the top dog suggests that talent and execution is incredibly important. A year ago none of my normie friends new them, and when i suggested using Claude looked at me like when I recommend Linux.
That shouldn’t affect Grok’ coding ability. How often are people discussing politics with Claude code? Writing decent code is just hard and it’s not just Grok.
If training a good model requires talent then that’s the answer to the question this thread is trying to answer: is training a good model actually that hard?
Talent to do.. what? This could mean a lot of things.
Navigating astronomically huge fundamentally not so hard but still really tangly and hairy projects requiring both excellent short- and long-term vision in an overheated domain with angry people and lots of money is a skill all of its own.
You’d be quite surprised, I think. Fine tuning a model on one axis can have drastic impacts on another that as a human we would expect to be completely unrelated.
If they have a top team and the money then appears to be a matter of a year or two? And one startup mentioned is Japanese not Chinese so they won't be banned from buying US tech.
My impression is that the answer is yes, that it purports to dispense the glue on-the-fly in some kind of dynamic way rather than being some kind of new model-amalgam.
And even if they didn't, they have a track record. Even if we did have benchmarks in this case I would still wait until people got there hands on it and formed a more holistic opinion.
No, stop right there. Anything published by Anthropic implicitly is not third party. For it to be third party, the third party has to be the one publishing it.
Just like many comments have been saying here, I also tested Fugu and some others and what I noticed is that they are quite expensive models, 20$ is not enough to complete a full workflow which in Opus it's possible, sure you might need to improve your prompt from the get go with Opus if you want the best results but so far that's my experience.
My next test will be Agentic systems and see how they perform
> These companies providing tokens, whether SOTA or not, that want to IPO are so fucked as time goes on.
>Can't sell their SOTA models, only slightly better than the open source models for the models they can sell, cost 20x to 50x for good models, a TAM that consists almost solely of developers, with no customer of theirs actually boasting increased profits as a result of AI...
> I fear their time to IPO may have passed.
What on earth could Anthropic and OpenAI Pivot to now?
I used to agree with you but now do not. I now think the floor for this market is probably no worse than the annual revenue of cell phone plans in the US market. So say, $250 billion.
Now, that probably doesn't justify the valuations and hype being thrown around, but I think it gets at a real revenue number.
I also don't know how that number fits into the funding rounds already raised and VC dreams of IPOs for these two.
This isn't coming from deep analysis on a verifiable source, but I started asking people in my social circle (includes white-collar and blue-collar folks) about their LLM use. The biggest surprise in 2026 for me was that almost all of these people told me about regular (and sometimes sophisticated) use.
A more intriguing observation - I work on the side with high school students and have two college kids of my own. Their LLM usage (and their peers) is much, much lower than expected . . . that's a little counterintuitive given "popular" perceptions I read.
> I used to agree with you but now do not. I now think the floor for this market is probably no worse than the annual revenue of cell phone plans in the US market. So say, $250 billion.
I don't think we're talking about the same thing. I'm talking about what their IPO is going to do to their share price.
In any case, $250b revenue translates to, best case scenario, $50b profit. On an investment of $1t. It does not look good for those companies making up the $1t investment.
Gotcha. I'm past the point of having any confident thoughts about what happens to their share price at IPO.
What about the idea that there is a high likelihood that the potential share price for OpenAI and Anthropic are both going to be pretty divorced from a rational market price for either?
Interesting idea and reasonable number, but cell phones need a lot of infrastructure and they need interconnection. The risk here is that in the future a combination of near-sota open weights models optimised to use as little resources as possible and a reasonable drop in compute price, will make possible for small and tiny providers to compete with Anthropic/ OpenAI or even for people to run their own private models for most applications. Then large, expensive sota models would only be used for research and to answer the small subset of general user queries that need that kind of intelligence.
The open models are half the equation. The other half is Apple's hardware, which is likely to see major memory bandwidth improvements over the next 2-3 generations and will be capable of running substantial models locally. By that point the open models will be beyond today's SOTA.
I agree with everything you said about their situation, but it's not like that is what will be evaluated in an IPO. There will be continued hype by the companies, lobbying to win support of a corrupt administration, and a narrative spin by clueless media about this AI revolution that will give investors fomo.
It’s hard not to be cynical as a dev. Every time I see a non-dev messing around with tooling I feel pain and discomfort.
You can bet your ass I can make AI make a consistent and repeatable powerpoint deck.
Problem is a) the need to produce this “deck” is questionable, at best and b) thinking in “powerpoint” is severely lobotomizing and limits you and your company in ways I find hard to even enumerate.
The exact same bullshit is going on with “word”. People can’t separate content from form apparently and that’s a debilitating disability no AI can fix. Creating AI to help you with “word” documents is an infinitude of stupidity so dark and so deep it makes my eyes water and fear for the future of humanity.
All I can say is imagine you do NOT have “powerpoint”. Let’s say all you have is a whiteboard with a 10% pen so you have to keep it short. What do you intend to communicate? Drill down and stop thinking in “powerpoint” and start thinking in content.
Content and form are orthogonal, repeat after me.
Once you have a stable and consistent information schema you can create “decks” till the cows come home and they’ll all be great. You won’t even need special AI for it in all likelihood. If you do not have a firm handle on the payload itself you’ll be forever tangling with “fonts” and “charts” getting none the wiser.
Oh man ya’ll are going to hate me so much so I’ll just spew further so I can get along with my day.
Imagine REST APIs being built around button designs, serialized CSS in business objects, in short imagine completely mishandling information and tanging it up with vague social indicators like “looks like he did work” and “this looks like an executive would be impressed”. Businesses are peddling proxies instead of raw information and for the love of God I can’t figure out why. AI or no AI this is a self-inflicted mutilation of such epic proportions I find it hard to imagine AI even budging the needle in such a low-density low-value environment.
Let’s face it - without the humans these machines ain’t shit - aka we have mechanically figured out ways to make machines better than us at certain things (on demand memory) but this idea they are intelligent is horse shit.
Btw the bar is low too! Most human created decks are garbage. And yet LLM’s don’t even beat those.
Machines have nothing to do with what speaks to the human - at best they are media to transmit what a human wants to express.
Having all the data in the world doesn’t solve that. Although if enough of society gets exposed to slop and there is an erosion of taste and quality… in fact I’m gonna stop there. It’s a dark path.
They may not get the valuation they want, but as it appears to be on a plateau may be better to offload now?
As per SpaceX, so many big names are involved the media will be controlled to hype it up and the investment banks will forecast 100x revenue in 2 years...
I doubt it will rival Mythos or the upcoming Sol, and if it's not open weights it doesn't really matter in the grand scheme of things. Still, I applaud the asian LLM efforts and hope they keep up the pressure on the americans.
if fugu really is an orchestrator dispatching to opus/gpt under the hood (as the openrouter page suggests), the $20-in-one-prompt complaints actually start making sense — you're paying api markup twice.
unless they launched 10t param models, or figured out some amazing new way to compress as many params into say 100b, I doubt it's anywhere near "mythos level". and I have no idea how many params mythos has but that was just some hear say.
Competition is accelerating, but the next breakthrough isn't just better models it's better connectivity. AgentKey bridges AI agents with real-world tools, APIs, and data.
It is not. Where's the danger ? We will need to adapt, as in every technology progress, but what do you think will happen ? Realistically ? Don't feed the fearmongering. Yes, we're disrupting the status quo, if that's the danger, then welcome to the world.
It's not one single danger, its a can full of dangers in various domains. And in contrast to other dangerous technologies, we are talking about one that has the potential to self improve. This smells like exponential growth doesn't it? Exponential growth is something we are not very likely to adapt to successfully, even if you say we are supposed to.
But before you complain, here are just a few concrete dangers, that come into my mind right now:
- mass layoffs in a system that is far from being prepared for sth like that. (no UBI)
- a Mr. Robot tier blackhat at the fingertips of every teenager in their mom's basement in a software landscape that is far from being prepared for sth like that. Side note: Big parts of the world including critical infrastructure runs on software.
- because it automates more and more intellectual work, it can cause mass brain atrophy, which isn't a hopeful sign for the human branch of the evolution
- increasing dependence on a technology, that is in the hands of those with capital.
And to the OP: this technology has potential implications that are far beyond 'being behind' other nation states economically.
Actually the real danger is mass labor market disruption, and a massive shift of power from labour to capital.
As was highlighted in previous discussions, the industrial revolution took 80 years to start benefiting workers. The continued impact of automation at least contributed to the rise of right wing extremism and an erosion of democracy all over the west. Now we face a development that has the potential to be faster than those that came before, in the context of political systems more fragile and worse equipped to manage the change.
So yeah, disrupting the status quo can absolutely be dangerous. It has been dangerous (and deadly) in the past and in the present.
> the industrial revolution took 80 years to start benefiting workers
Come on. This is dishonesty and isn't the reality. We may agree that the Industrial Revolution may have taken decades (certainly not 80 years) for its benefits to be *clearly and widely* felt by workers, but anything further is an abusive claim. So what, because the progress doesn't benefit to workers instantly, we shouldn't do it ?
In the end, whatever your position, industrialization eventually raised living standards. So what's wrong with that ?
> The continued impact of automation at least contributed to the rise of right wing extremism and an erosion of democracy all over the west
This is oversimplifying and correlation at the best, not causation
"Carl Benedikt Frey at Oxford has documented that the Industrial Revolution took seventy years before wages and employment recovered for the workers it displaced. In the interim, wages stagnated, the labor share of income collapsed, profits surged, inequality skyrocketed, and the political consequences included the Chartist movement and widespread social upheaval."
From half way through this (meandering) blog post:
Any superintelligence operating under a consistent moral framework will decide to extinguish humanity with as little ecological damage as possible, because humans cannot coexist with other life on this planet. It will realize that a bioweapon is the ideal choice.
> Any superintelligence operating under a consistent moral framework will decide to extinguish humanity with as little ecological damage as possible, because humans cannot coexist with other life on this planet.
There are plenty of internally consistent moral frameworks which would not favor this action even if the premise were true (and that premise seems at best unjustified and at least overstated.)
I think it is time that we had a UN-sponsored standards body dedicated to bench-marking the newest models from around the world, for everyone's benefit.
Given the national security implications, it's no surprise that Japan and China are rushing to build sovereign models post-ban.
But when these startups claim parity with "Mythos," could it be that they are just optimizing for very specific inference tasks?
I wonder if we are seeing the real battleground shift from raw training scale toward specialized inference.
asian is bad wording. this is a japanese startup backed by khosla ventures. japan is an ally of west.
the title makes it sound like a chinese company did this.
Are you from UK perhaps? Because UK is the only place I know where "asian" also includes India and Pakistan. Everywhere else India is sort of separate entity and always mentioned separately.
You don't get out much? (in alpha) Asia, Australia, Canada, Europe, Middle-East (sorry, that's a western term) all include west Asia in the definition.
I'm from Europe and never hear indians clumped together with chinese as "asian". When speaking of geography, then yes, but that's about it. It's always "Indian startup", never "Asian startup" for example. Indian food, not Asian food etc. YMMV
The UK is part of Europe despite leaving the EU (which is of course not equivalent to "Europe").
In the UK it is common for the British to refer to both Pakistanis and Indians as Asian and not uncommon for the Chinese to be lumped in the Asian category as an umbrella term (and also to highlight someone or some company as being specifically Chinese).
I guess it really comes down to habits in particular parts of Europe and which parts of Europe you travel in.
Is that really the most sailent facet of this story? Boxing it by official friend vs foe designations? Don't american academic institutions and corporate entities cooperate closely with Chinese companies as well?
The US and China are in a cold war right now, whether that is fully recognized or not, the fight has already begun. The US is blocking models from getting out of the country and China is blocking researchers from getting out of the country. The expectation should be only more closing off in the future.
We are all people. This ally-of-the-west framing is propaganda. Who has harmed me more: this US or China? Who do I have more in common with: a tech worker in China or a US government official?
(I'm based in US - I use the best tech for the task).
I seriously do not comprehend how a consumer like you can have sympathy for Anthropic, as if you are part of their organisation or something. Competition is good for us. Wouldn't it have been for asian labs, we would would be fully dependent on OpenAI, Anthropic and Googles services.
Anthropic just stole the internet and put it in a transformer and pat itself on the back for it - well no to be honest we have to suffer through hearing them saying that this model is really really dangerous until they got a reaction for they fear mongering
Anthropic are the pathetic ones. The pariah of the AI industry that nobody likes because all they do is lie, cheat and steal. Now no one can access ChatGPT 5.6 because of their 5 year long fearmongering regulatory capture campaign.
Hopefully they go bankrupt and someone else takes their place.
> Now no one can access ChatGPT 5.6 because of their 5 year long fearmongering regulatory capture campaign.
I'm sympathetic to this arguement, but it's silly to ignore the other half; that the administration has openly feuding with them for months over limits to military capabilities.
No one is ignoring the other half, the feud is rooted in Anthropic's insatiable desire for power and control over everyone and everything, including the administration. The same desire that is fueling the strategic fearmongering campaign underpins all of their behavior and the repercussions and sentiment they're facing from the administration and the general public.
If their company hadn't been posturing like this for 5 years they'd have played ball with the administration like all of the other AI companies and they wouldn't have caught all that heat and taken down the AI industry with them. Just remember that Dario was pushing the narrative that GPT 2 was too dangerous to release to the public, while he was working at OpenAI. GPT 2!
Now it's an inevitability that China takes the lead - which was probably the case anyway, but a certainty if this continues.
> No one is ignoring the other half, the feud is rooted in Anthropic's insatiable desire for power and control over everyone and everything, including the administration.
Again, this is ignoring half of it. See what they did to Intel prior:
"President Donald Trump said Monday that he and members of his cabinet met with Intel CEO Lip-Bu Tan, days after he called on the head of the chipmaker to resign. Intel shares rose 2% in extended trading."
"U.S. Government to make $8.9 billion investment in Intel common stock as company builds upon its more than $100 billion expansion of resilient semiconductor supply chain."
Do what you wish with this info, but it seems to be a complete waste of $$.
reply