The restrictions on LLM models like Fable has created a viable international LLM market where it was difficult to justify investment two weeks ago.
As a non-US citizen Opus 4.8 is the best American LLM I will ever have access to. That's no longer up for debate or question. Each month that I pay Anthropic is now a depreciating value -- I'm paying for models I'll never be able to access, while other models are able to catch up.
Adding US based identity verification through Persona is also incredibly off-putting. I think it's sufficient to kill my use of Claude altogether.
So the question I have to live with is what do I do instead.
I installed Mistral Vibe last week and I've been experimenting with offloading work to it. I won't pretend that Mistral-medium is close to state of the art. It isn't. It still writes incorrect tool calls.
From the last week about 50% of my LLM tasks actually reduced to "take this work and write about it" and Mistral excels there -- it definitely beats Opus at writing. Mistral nails it, and when it doesn't its so fast to iterate.
There's another say 30% of tasks that's writing queries against a data warehouse. I updated my semantic layer MCPs and Vibe uses them, but it struggles with ambiguity here. It's not a replacement, it's maybe where Opus was a year ago.
The rest of my work involves writing code. That's going to be harder to replace for now. My next step is exploring OpenRouter and other models. I can't decide if I was ever actually happy with Opus's work on this front though -- the understanding tradeoffs when you trust LLMs with decisions stack non-linearly and negatively. I did like Fable on these tasks, I won't lie, I will miss it, but not by any choice of my own.
> As a non-US citizen Opus 4.8 is the best American LLM I will ever have access to. That's no longer up for debate or question.
This is a crazy conclusion for a situation that isn’t even two weeks old. LLMs are not the first tech product that have been restricted by export controls. These situations pass. Administrations change. Technology evolves. We’ve had export restrictions on different chips and even cryptography software in the past. It doesn’t last forever.
Sincerely or not, judgement is yours, Dario has been begging for regulation. He has been talking about how Claude models are distilled by foreign adversaries. And now the regulation is here.
What makes you think this situation that the CEO of Anthropic is asking for it temporary? Do you not believe Dario was sincere?
1. Identity verification as a way to validate real-personness and mitigate distillation by e.g. North Korea
2. Identity veriiication as a way to limit model usage to US residents / citizens
3. The level of model which will be subject to identity verification, today and as time goes on
It’s a mistake to conflate the three and form a rock solid opinion of exactly what will happen from here to the heat death of the universe. Everything about AI is moving quickly. I doubt Dario would claim to have a perfect roadmap not subject to change.
My personal guess is that just like export controls on CPUs, this will apply differently to different regions, and will change over time. Especially with US political instability and increasing anti-science policy, I cannot imagine Dario or anyone else would want to surrender the EU and other markets to become a US-only company.
But whether I’m right or wrong, one thing I’m not is certain. I can’t imagine how anyone could be in the current situation.
I can't speak to your specific use of Anthropic's models, but I find it interesting that people will identify themselves (to set up and pay for an account) and provide all sorts of personal (and often sensitive if not confidential) information to these models on a daily basis, but balk at a 5-minute identity verification.
The big problem American policy makers (and business leaders) don't understand is that they tend to minimize or ridicule extremely serious events.
There's a pre-Greenland and post-Greenland annexation threats for European Nato allies, and it is non reversible. EU allies do not forget that the US (the only country to ever call article 5 or to gather NATO allies for operations) has both mistreated the alliance, and has been the only power to threaten militarily EU countries.
Same happens here. Business-level wise, you seem to be talking with a very American-centric point-of-view, like these events are minor and temporary issues and we're all here waiting to throw money at an abusive relationship.
But this is not how we operate in EU. None of us can afford to build their operations based on uncertainty of US export controls. The damage is here and many of us are replacing Claude/GPT subscriptions with shared opencode servers using GLM and DS4.
Might be slightly worse? Probably. But we can work on it, harness it, get experience, and even update back to American models at some point. But we're no longer going to be building assuming US models availability.
If it happens once that means it can happen which means it can happen at any time.
When you feel you're a 2nd class person, or 'other', you're not eager to empower your oppressor, quite the opposite.
Completely agree.
I mean...use opus/fable when you can, if down the road your access gets cut then just switch to kimi or whatever.
Yeah, this sucks, but you're being really dramatic and acting like you can't switch llms with basically no lock in. Getting something like your email cut off would be a real thing to be concerned about, but this isn't that.
Before this incident I’d gladly use any anthropic LLM in production features. Right now, this has become a risky decision that can tank my business overnight.
If your business-critical systems rely on SAAS that doesn't have a solid SLA and breach-of-contract provisions that more than cover the damages in the event, you've made "a risky decision that can tank [your] business overnight".
If the software your business depends on can't run indefinitely without getting permission to operate from someone else's systems, then you're perpetually at risk of someone else tanking your business because they decided that you can no longer use that software.
But not on a SaaS whose continued availability you'd rely on.
In any case, your optimism is bordering on naivety. The world has seen how the US can easily disregard anything and act arbitrarily - sanctions, tariffs, shutting down access to SaaSes - and this will not be forgotten. As you say, administrations change. Even if next time around there are competent adults in the White House (which really isn't a given), do you really want to bet your business on that not changing 4 years later?
There's a reason why all the big cloud providers are constantly shouting about their "sovereign" solutions. The US has broken everyone's trust and there is no going back on that.
I think, this is all a culmination of rapidly eroding trust and soft power between US & its allies for the past 3y.
You threatened to invade Canada and Greenland.
You surely don't think you're coming back from that?
Well, irrespective over as to whether this is the case, the blog entry from claude came yesterday, aka June:
https://support.claude.com/en/articles/14328960-identity-ver...
So, why the two months delay here? If they felt all was already said, they would not have had a need to repeat what they wrote two months ago already that mandatory age sniffing is required for all claude users.
Yeah, in the long everything will happen, from 1,2,3,4,5, and 6 being the winning Powerball numbers to heat death of the universe, but as the colloquial goes "ain't nobody got time for that!"
Once an institution or person has proven that they will take adverse action against you, it is foolish to bank on them again.
This is off-putting to the HN demographic, but won't change anything in practice. 99+% of people will just do the ID verification and move on.
- https://cybernews.com/privacy/persona-leak-exposes-global-su...
- https://hothardware.com/news/discord-drops-persona-after-use...
It’s off putting outside of HN
Why? Is Claude really so much better that the additional hassle and privacy invasion is worth it? What's stopping people from switching to one of the dozen or so other AI tools?
Heck, considering the volatility of the LLM industry, shouldn't everyone already be using OpenRouter & friends to avoid getting screwed over by the model-of-the-week - making a switch absolutely trivial?
I’m lazy.
I will simply upload my drivers license to Claude, and continue paying $200/month.
A lot of people are already skeptical of the frontier labs - I moved over to Claude from OpenAI when they bent over for the US government. And I'll certainly move on again if Claude start asking for photo ID.
In a vacuum yes. But this space isn't vacuum, I'm going with path of least resistance.
> Each month that I pay Anthropic is now a depreciating value -- I'm paying for models I'll never be able to access, while other models are able to catch up.
Excellent point... That made me rethink my payments to Anthropic. As one of the foreign peasants who was banned from accessing Fable by the land of the free, it's become really hard to justify giving Anthropic any more money. I'm very tempted to switch to GLM 5.2.
It should go without saying, but "the free" in "the land of the free" refers to Americans, not everyone else. Not sure if you were trying to make it sound like a great irony that "the land of the free" would exclude or ban people, but if so it doesn't quite hit the mark. A more cutting criticism would be that the land of the free isn't letting one of its own companies freely compete.
I'm playing with Deepseek a lot more via OpenRouter recently and the only major downside I can see is the usage billing over the monthly plan
The monthly subscription is also a major hurdle for me. The "high end frontier models for low prices" aspect is a major reason. I think I'm getting a lot of value from my subscription, given that the API prices are like a hundred times higher.
However, there's also a psychological factor here. These subscriptions are like the gachas of the software world. I got "addicted" to them. I developed workflows around achieving 100% weekly usage. Sometimes Anthropic randomly resets weekly usage and I scramble to get the most out of it. I'll point Claude at things and then just have it run hundreds of code review agents. I ran out of projects to do this on and started doing it to my favorite open source projects instead, looking for things to contribute.
I think with usage-based pricing I wouldn't use LLMs quite so freely. It'd probably cure my "addiction" too, but the problem is I'm not sure whether that's a good thing, since this "addiction" has been a somewhat positive force in my life. It's driven me to start new projects and also make major progress on existing ones. It brought me out of a slump. I'm a little afraid of moving away from Claude and not being as driven as I was before.
That’s totally unclear. Things are changing fast. No statement from god or potus has come down about the future of LLMs and who can access what. And for what it’s worth, I’m not able to access fable and I’m a US citizen.
Lol, welcome to US foreign policy. Or US trade policy. Totally unclear, and it's a feature.
Yes, lack of stability and clarity is entirely why people are steering away from the US. Yes things can change again next week, it's not a good thing.
Being unclear is enough for people to steer clear of it...
If you have a person as president that changes his opinions faster and more often than their underwear, you're simply not reliable.
At this point the future availability of Anthropic models outside the US is very unclear.
(Read: the US lacks authority to ban use by citizens and doesn't want to risk their hand in court, particularly since lawyers who know anything have all left US government and what remains are complete incompetent jokes who can't even win slam-dunk cases due to repeated procedural errors. The nice thing about blocking non-citizens is they are easy to bounce out of court on standing)
I would suprised if admin doesn't want american companies and their employees to not be over competitive with outside companies.
But I do see them wanting a lever to prevent international rivals from having it.
Of course, as you say, this would require govt funds to fix. Congress gets asked for funding to save the US economy, and various members of the Trump dynasty pop up in unexpected non-exec directorships for companies receiving those funds.
The problem with the ID verification is that they can pair introspective conversations with ID. Either that bothers people or it doesn’t.
Main point: we can’t fret about current state models because the ID verification has future implications. Models will change and competition will catch up. Do what feels right in the long run not whether TODAYS model is better at Anthropic.
I really wish this would change soon but they are not there yet.
On NeuralWatt for my personal projects at home (not affiliated, just a happy customer), I get so much more mileage out of GLM than I get out of Claude at work, specifically because it's priced as a hammer I can pound any nail-shaped-object with, not a delicacy I need to carefully budget-analyze to try to figure out if it's worth burning my monthly spend limits on this task.
https://openrouter.ai/compare/z-ai/glm-5.2/anthropic/claude-...
Note the big cut in token prices from China.
After the recent 75%+ price cuts by DeepSeek & Xiaomi MiMo (and now, MiniMax), I pretty much packed my Claude bag up and moved over. I see no discernable difference (other than chattiness of its thinking modes) in capabilities for the kind of coding & debugging work I do.
Then GLM came out and that just means everything got even better.
They will flip to being just as bad or worse if they beat America.
But America isn’t deserving trust at this point.
The only viable option now is local AI. Our industry needs to figure out how to decentralize training data, infrastructure, inference and analysis.
Not only is it up for debate, I find it extremely unlikely to be true.
I don't really disagree with the rest of your post but I very strongly doubt that Opus 4.8 will be the best American LLM you'll have access to this year.
It would be extraordinary if this was the beginning of the first ever permanent technology export restriction.
Even if it's never lifted, probably GPT 5.6 will be better than Opus 4.8 and they won't hype it up as too dangerous to release before releasing it.
Open models will catch up to Opus 4.8 and state of the art will, most likely, continue to get better. It won't be long before "better than Opus 4.8" is not a big deal.
Your overall argument makes some sense, but I would bet any money this simply is not true. Even if the US maintains some of the restrictions on export (which is in no way a given with how fickle this administration is, let alone the next administration), as LLMs advance, Fable will eventually be considered a lower tier model, and is likely to have restrictions lifted.
Your alternative:
Pay a Chinese or EU company for a third rate model.
The choice is yours.
> The restrictions on LLM models like Fable has created a viable international LLM market
The issue is that there is no "international LLM market." America is leading the AI race, and while Chinese open weight models are great, they aren't quite bleeding edge. I routinely use Qwen and GPT-OSS (locally) for things I don't want to share with Anthropic, but they are clearly inferior to SOTA cloud models.
this sentiment is far too commonplace in my opinion.
just because the BMW x5 exists doesn’t mean rav4 isn’t used by far far far more people.
the rav4 is close enough for most people to the x5 and way cheaper.
if deepseek is close enough and significantly cheaper, which direction do we think the market will go once the hype trash moves on and people realize how much of the hype trash is just botted algorithms?
Codex-5.5 > Opus 4.8, so that's not true.
Some people have been saying that 5.5 is on par with Fable and it's just nonsense.
Definitionally, slightly better than China should always be fine for export.
You think some people aren't "deserving" of the best models?
That is very well put.
Now everyone talks about Fable and wants Fable.
Having used it for limited time when it was available, I don't miss it at all.
So I don't even know if I miss it. I suppose that's equivalent to never having used Fable at all.
This identity verification is a best effort to kinda stay afloat: they can now offer bleeding edge models to US nationals, but not to the other 95% of the world. Their influence is gonna tank quite seriously if the previous mandate is not reversed.
Realistically, their choices are to either implement this, or restrict access to new models entirely, which is a sure way to fall into complete irrelevance.
Worse, doing this forces other nations to catch up to compete. Once they have, what happens next? An AI arms race that the US may not win? Someone else opens their Fable class model first and takes the multi-trillion dollar market that could have been run by a US company?
There is no n-step positive outcome for the USA. The only winning move was not to play.
...
> My next step is exploring OpenRouter and other models
May I suggest using Cortecs.ai then? OpenRouter is US-based as well and since you have been bitten by this already perhaps it's really time to change course? :)
I'm starting to see more and more of this: speed matters more than model, skills matter more than model, cost matters more, harness matters more than model. It seems like until we have a step change in models (and Fable isn't it), there's a lot of room to optimise with what we already have.
This is a stark change to the "best model at all costs" mentality from a year ago.
Not exactly. The close buddy with the owners of several other AI companies who has a known dispute with Anthropic and who has seized near-absolute power in the US is doing something that damages Anthropic. The fact that it also destroys long term prospects for the US overall is irrelevant because the individual can't think ahead and also doesn't care about the future of the US.
It's nearly the same thing in the end, but helpful to understand the cause.
I’m really enjoying the phone-controlled sessions with Claude, and the models do good work; but if I’m asked to validate my ID, we’re done.
I already have Aivo and Pi somewhat similarly configured. Just give me a reason and I’m giving my money to someone else. In Europe, if you have my ID it’s not that hard to get me into trouble. And there’s no safeguard that Persona won’t be hacked, it’s not like they encrypt the ID with my own keys or something.
They do however want strict measures in place to avoid abuse, and the export control was the only tool they had to stop Anthropic from releasing the model.
Though I also wonder if it's even possible to patch things without severely crippling the abilities of the model.
And then for stuff that you already said you were able to use Mistral for... Qwen-3.6 (option to run locally), MiMo-2.5, DeepSeek-V4-Flash, or... many, many other models to choose from too.
either this or stick with Codex until the US government cripples it too
No, we are all just waiting for Dario to get scheduled for an Oval Office press conference where he can present a gold trophy to Liberace Hitler and extoll his praises for all the amazing winning he is doing like no one has ever seen before.
(Disclaimer: I work for Databricks, but do not work on omnigent - though I have submitted some QOL PRs as a community member)
