Hacker News Reader

Google and OpenAI have done similar things with their LLM offerings.

Gemini Advanced offered 2.5 Pro with nearly unlimited rate limits, then nerfed it to 100/day.

OpenAI silently nerfed the maximum context window of reasoning models in their Pro plan.

Accompanying the nerf is usually a psy op, like nerfing to 50/day then increasing it to 100/day so the anchoring effect reduces the grievance.

It's a smart ploy because as much as we like to say there's no moat, the user does face provider switching costs (time and effort), which serves as a mini-moat for status quo provider.

So providers have an incentive to rope people in with a loss leader, and then rug pull once they gained market share. Maybe 40% of the top 5% of Claude users are now too accustomed to their Claude-based workflows, and inertia will keep them as customers, but now they're using the more expensive API instead. Anthropic won.

Modern bait and switch, although done intelligently so no laws are broken.

It’s not just switching costs though, what alternatives to Claude Code are as good right now?

I had been loyal, but am not any longer, so the ploy definitely did not work with me. I guess I'll move on to Gemini now, until I get sick of it.

To the degree there is a moat, I do not think it will be effective at keeping people in. I had already been somewhat disillusioned with the AI hype, but now I am also disillusioned with the company who I thought was the best actor in the space. I am happy that there is unlikely to be a dominant single winner like there was for web search or for operating systems. That is, unless there's a significant technological jump, rather than the same gradual improvement that all the AI companies are making.

> I had already been somewhat disillusioned with the AI hype, but now I am also disillusioned with the company

Likewise: a faulty, unproven, hallucinating, error-prone service, however good, was a good value at approx 25 USD/month in an "absolutely all you can eat", wholesale regime ...

... now? Reputational risk aside, they force their users to appraise their offering in terms of actual value offered, in the market.-

> they force their users to appraise their offering in terms of actual value offered

That's a good thing, right?

Very much, absolutely, most resolutely and definitely, yes.-

Loyalty is for henchmen.

I am loyal to my family and friends, business partners, and to those who I believe are fighting the good fight, as long as they continue the good fight. I do not consider myself a henchmen to any of them.

Loyalty is a conflict resolution tool. But when my relationships are such that nobody is pursuing contradictory goals to begin with, I find that it doesn't come up hardly at all.

On the rare occasion that it does, I try to circle back and mitigate the root cause so that I can resume a loyalty-free life thereafter.

You probably won't get sick of Gemini. Your wallet may need to get used to an adjustment though.

Interesting to see the definition of moat change from keeping other companies out to keeping your customers in.

> the user does face provider switching costs (time and effort), which serves as a mini-moat for status quo provider.

When a provider gets memory working well, I expect them to use this to be a huge moat - ie. they won't let you migrate the memories, because rather than being human readable words they'll be unintelligible vectors.

I imagine they'll do the same via API so that the network has a memory of all previous requests for the same user.

Unintelligible vectors might be hard to transfer from one of their older models to one of their newer models - so I think the human readable words will remain a bit of a narrow waist in this space for the immediate future at least.

Gemini 2.5 Pro already uses encrypted 'thoughts' in the context window which are not visible to the user. They might be English words, or might be some other internal state vector.

Is memory all that useful for using these LLMs? I’ve found that I mostly use them for discrete tasks - helping me learn a specific thing, build a specific project, debug a specific piece of code, and once it’s done I’d actually prefer it to forget that thing instead of keeping it around forever.

Hell, “just open a new chat and start over” is an important tool in the toolbox when using these models. I can’t imagine a more frustrating experience than opening a new chat to try something from scratch only for it to reply based on the previous prompt that I messed up.

I wonder if that's actually illegal, because it feels very close to false ads etc. It seems legal, but I think courts would side with customers.

As if google would say that yes, emails are $5/mo, but there's actually a limit on number of emails daily, and also number of characters in the email. It just feels so illegal to nerf a product that much.

Same with AI companies changing routing and making models dumber from time to time.

If there is a material difference in the product that causes you to no longer feel that it's the same as it was when you subscribed, it could be considered a bait and switch? I think as soon as you notice that this is the case, you should probably stop paying them though, otherwise you might seem to accept this state of affairs. If you had a long term contract that didn't have some kind of language that tried to prevent this from happening in the first place, you could probably get out from under that contract by saying that the deal has essentially changed out from under you, but I think a lawyer might make that argument much better than me.

I'm not sure what harm you think you're suffering from, and what a proper remedy might be, if you think it's illegal. I don't know if I would go that far, as there are all kinds of words most terms of service use to somehow make it so that you have already acknowledged and agreed to whatever they decide to do. So a lawyer will probably be helpful there as well.

I don't even want to imagine how bad it will get if a legitimate moat does surface and users get entrenched ever deeper into the ecosystem of a single provider. A lot of the companies in this space have a track record of squeezing as much value out of their customers as they can get away with.

Google tried to offer a free tier for students and people to TRY it. That's a LOT different because they were up front about limits.

You pay for Gemini by the token and you get the full firehose. It costs money, but less than Opus and it smokes that.

It just works. Gemini 2.5 Pro is the king of AI coding and literally everything else has to catch up.

Trust me, I can't wait until there's a model that can run locally that's as good...but for now there isn't.

Always just look at the token cost and get used the token economics. Go into it paying. You'll get better results. I think people thinking they were somehow cheating and getting away with something similar (or better) for $20/mo are in for a big surprise.

I don't know if I would say they should have known better of course. I think Anthropic and Cursor and Windsurf were hiding it a bit. Now it's all coming out into the open and I guess you know the saying, if it's too good to be true...

Claude definitely engage in some shady and illegal behavior. I wanted to see what the annual plan would cost as it was just displaying €170+VAT, and when I clicked the upgrade button to find out (I checked everywhere on the page) then I was automatically subscribed without any confirmation and without ever seeing the final price before the transaction was completed. Turns out that price was €206.50. Getting refunded was also a pain in the ass.

You might not have noticed in all the frustration, but you got overcharged. 170 + 20% VAT is 204, not 206.50.

Maybe they added a card fee in at the end, but if they didn’t make that abundantly clear, they’ve broken a law in most countries which use the Euro.

Why 20% VAT? Different EU countries have different tax brackets, so OP might not have had 20%.

There’s a European country with a 21.470588235% tax rate? No I think GP is right and there’s a hidden 2.50 handling fee.

Lol, that's a fair point, you're probably right on that.

Probably because VAT_floor(206.5/170)=20

VAT is not 20% everywhere worldwide.

€170 + 21.5% (Irish VAT rate) is €206.55. So not sure what you expected. One complain though in EU B2C companies must show prices with VAT included.

By clicking, he expected to see the full price with VAT included, as required by EU regulations, without doing the math himself.

> €170 + 21.5% (Irish VAT rate) is €206.55. So not sure what you expected.

Parent clearly stated they only saw "€170+VAT" and not €206.55, so of course they expected to see €206.55 before the purchase went through. Not sure what anyone else would expect?

Even the ex VAT price is wrong. It should always just be 206.55 - can't tease low price first ex tax.

Why would anyone complain about showing prices with taxes (VAT) included? It's deceptive otherwise.

Went to see if I could cancel my Claude pro plan just now. On billing I see adjust plan button. On that page I can upgrade but no where can I find downgrade/cancel. I checked the account page. I see delete account but "To delete your account, please cancel your Claude Pro subscription first."

Update: below the fold at the bottom of the Billing page is the cancel section and cancel button.

Update 2: just clicked cancel and was offered a promo of 20% off for three months...

Update 3: FYI, I logged in to my Claude account via computer (not iOS or Android).

i tried this to see if i could get a discount and it just canceled it, lol. i wonder if its bc im a heavier user?

My credit card web portal has a section for reversing transactions without having to call anyone. Maybe yours does too?

Reversing transactions is not cancelling the service I believe?

As long as you don't cancel, you do owe them money. But if they make cancelling intentionally hard, one would likely have a good case in court to still not pay, if one would want to go to court over this.

Document the failed unsub process so if they bother to sue you then you win:)

nice feature. so its like backcharge via app instead of calling credit card customer care. do you mind sharing the credit card company

This is getting to be the norm rather than a unique feature. I can dispute a credit card transaction through the app for my Citi, X1, or Fidelity cards.

Tangerine Mastercard Canada.

My bank app has a bit where it detects common services and tells you how to cancel them in the app, like Amazon Prime

I've now got it behind OpenRouter on Bedrock and Vertex and am rotating in K2 and Qwen more and more as I learn how to get the most of them because the silent degredation and work avoidance heuristics are not a variable I need in an already difficult expectation management regime to my stakeholders.

At the rate the Chinese are going it won't be long before I can shake the dust off my sandals of this bullshit for good.

Qwen3 coder is very good and a tenth of clcaude price on openrouter. And has a :free version for light usage.

I still revert to gemini pro 2.5 here and there and claude for specific demanding tasks, but bulk token go trough open weight model at the moment.

i use privacy dot com cards to prevent this issue, takes 2 seconds to turn off the card in my extension