Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Anthropic downgraded cache TTL on March 6th

github.com

Post reply on HN
Print view View original

411–420 of 447 posts

Re: Anthropic downgraded cache TTL on March 6th

#411

Post by alphabettsy »

FireBeyond wrote: post #373
jclardy wrote: post #276

Earlier quoted context omitted.

Just anecdotal, but I was using Claude Code for everything a few months ago, and it seemed great. Now, it is making a ton of mistakes, doing the wrong thing, misunderstanding context, and just generally being unusable. I now have been using Codex and everything has been great (I still swap back and forth but generally to check things out.) My theory is just that the models are great after release to get people switch…

Yeah, shorter time frame but I've been noticing that too. Just the other day I was experimenting with some workflow stuff. "Do x and y and run tests and then merge into develop." Duly runs, and finishes. "All merged into develop". I do some other work, don't see any of this, double check myself, I'm working off of develop. "Hey, where is this work?" "It is in this branch and this worktree, as you would expect, you wi…

Prompt cache expired?

Re: Anthropic downgraded cache TTL on March 6th

#412

Post by alphabettsy »

beering wrote: post #233
matheusmoreira wrote: post #45

Earlier quoted context omitted.

I certainly noticed a significant drop in reasoning power at some point after I subscribed to Claude. Since then I've applied all sorts of fixes that range from disabling adaptive thinking to maxing out thinking tokens to patching system prompts with an ad-hoc shell script from a gist. Even after all this, Opus will still sometimes go round and round in illogical circles, self-correcting constantly with the telltale…

The real question I see nobody asking is how GPT-5.4 beats Opus at a fraction of the price. I doubt it’s only a question of subsidization. My impression from the past is that GPT-5 was around a Sonnet-sized model, and 5-mini was Haiku-sized. At least on my codebase anyways, Codex one-shots tricky things that Opus needs several tries to fully get right.

IMO it doesn’t handily beat it.

It’s typically equivalent, sometimes better, sometimes behind. Better at following a well defined plan, less good at concept exploration and planning imo.

At 1m context it’s basically the same price.

Re: Anthropic downgraded cache TTL on March 6th

#413

Post by siva7 »

brookst wrote: post #299
ledauphin wrote: post #144

Earlier quoted context omitted.

nah this doesn't explain it. most of the users of those third party harnesses care just as much about hitting cache and getting more usage.

Sure. The question is whether they have the same level of expertise and prioritization that Anthropic does.

They are working with the same tools and knowledge like Anthropic does as Caching practices are documented. And they have as much incentive as Anthropic does to not waste compute. Can we stop acting like people who build harnesses be it Opencode oder Mario Zechners Pi are dumbfucks who don't understand caching?

Re: Anthropic downgraded cache TTL on March 6th

#414

Post by jwitthuhn »

TheTaytay wrote: post #335
benced wrote: post #221

Anthropic responded: https://github.com/anthropics/claude-code/issues/46829#issue...

This should be the top comment. The OP misunderstands the change and has their LLM write an expose. The company responds with a well-reasoned explanation that it would actually cost MORE money if there was a global 1h default for ALL prompts. It gets downvoted and the pitchforks stay out because…I presume the words like “cache read likelihood” sounds like made up fluff to the audience, rather than an actual explanati…

It only potentially saves money for people on API pricing, it exhausts tokens faster with no benefit for users on the Claude Code subscription. Those users had their cache TTL reduced from 1 hour to 5 minutes and are saving no money because they were not paying based on the cache time in the first place.

Re: Anthropic downgraded cache TTL on March 6th

#415

Post by pllbnk »

supermdguy wrote: post #234
benced wrote: post #221

Anthropic responded: https://github.com/anthropics/claude-code/issues/46829#issue...

Bizarre reading the thread, it feels like their Claude responding to the other posters’ Claudes

It feels (nobody can prove it) that all user-facing applications are fully vibe-coded and no internal developers have any idea how they work, so they just keep redirecting user questions to Claude to answer on behalf of them. That's why they are dealing with regressions and downtimes every few releases as it's the usual pattern with vibe coding that bug keep resurfacing.

Re: Anthropic downgraded cache TTL on March 6th

#416

Post by onoesworkacct »

zoogeny wrote: post #370

As an aside, I built a tool to manage my own chat interface over the provider APIs. I added caching because the savings are quite significant and I have a little countdown timer that shows me how much time remaining until the cache is expired. However, for the basic turn-based conversation the cache (at 5 minutes) is almost always insufficient. By the time I read the LLM response, consider my next question, write it…

give it a skill that runs a timer in the background and every 4.5 minutes says "ping? pong!"

Re: Anthropic downgraded cache TTL on March 6th

#417

Post by sigseg1v »

deaux wrote: post #218
cedws wrote: post #193

Earlier quoted context omitted.

When you're using the SDK, yes it can. Example: I used the Python SDK to translate a bunch of source code recently. I spawned a subagent for each module that needed translating and left it to run for a few hours with a parallelism limit of 5. It blasted through the 5 hour usage and dug into extra usage credits. I have zero assurances that the above can't result in a ban. The usage pattern is not distinct from OpenCla…

As I said, it doesn't just happen, you explicitly had to set it up so it could happen.

Just in case it wasn't clear, what they described doesn't need extra tooling. You can write this in your CLI and it will easily cap a Max 20x plan in an hour: "we are converting this entire codebase from TS to C#. Following the guidelines I've written in MIGRATION.md, convert each file individually. Use up to 32 parallel subagents. Track your work for each file in a PROGRESS.md file, which you will update for each file starting and completing. Using an agent team, as a secondary step, add a verification layer where you verify each file individually for accurate migration following the instructions in VERIFICATION.md"

Yea there are other ways to do this, you can set up a separate harness sure to make it more efficient, but just the above will also work, it's just text you paste into your CC terminal, and it will absolutely cap the largest subscription plan available no problem.

Re: Anthropic downgraded cache TTL on March 6th

#418

Post by theshrike79 »

disillusioned wrote: post #8

It's also routinely failing the car wash question across all models now, which wasn't the case a month ago. :-/ Seeing some things about how the effort selector isn't working as intended necessarily and the model is regressing in other ways: over-emphasizing how "difficult" a problem is to solve and choosing to avoid it because of the "time" it would take, but quoted in human effort, or suggesting the "easier" path f…

Am I the only one who couldn't care less if a model can answer a weird gotcha riddle or not?

I never use it to answer questions like that, what I care about is consistent tool callig and following the prompt.

Re: Anthropic downgraded cache TTL on March 6th

#419

Post by sigseg1v »

shimman wrote: post #375
skullone wrote: post #360

Earlier quoted context omitted.

I don't know about "only doing trivial things". I've built a fully threaded webmail replacement for Gmail using imap, indexes mail to postgres, local Django webapp renders everything in a Gmail/Outlook style threaded view with text/html bodies and attachments and a better local search than gmail, and runs all locally. Started as a "could I?" and ended up exceeding all my expectations

That would be considered a trivial thing, why wouldn't it be? It's just basic crud you're doing. Nothing unique and that hasn't been written about tens of thousands of times across millions of books/blogs/comments before.

I used it to analyze a single player game binary from steam, hook into all the relevant game state modifying functions, add full multiplayer state sync and then also hook all the relevant portions of the UI to add multiplayer to the game.

I'm not saying it's the hardest thing but I also wouldn't consider it trivial.

Re: Anthropic downgraded cache TTL on March 6th

#420

Post by mrgill »

andai wrote: post #65
sunaurus wrote: post #16

Has anybody else noticed a pretty significant shift in sentiment when discussing Claude/Codex with other engineers since even just a few months ago? Specifically because of the secret/hidden nature of these changes. I keep getting the sense that people feel like they have no idea if they are getting the product that they originally paid for, or something much weaker, and this sentiment seems to be constantly spreadin…

Well, off the top of my head: - Banning OpenClaw users (within their rights, of course, but bad optics) - Banning 3rd party harnesses in general (ditto) (claude -p still works on the sub but I get the feeling like if I actually use it, I'll get my Anthropic acct. nuked. Would be great to get some clarity on this. If I invoke it from my Telegram bot, is that an unauthorized 3rd party harness?) - Lowering reasoning eff…

For what it's worth. I invoked claude -p from a script, and my account was nuked immediately. DM'd Thariq from Anthropic who admitted it was a weird classifier and would look into it, but then he never followed up. Been 13 days since I've been banned now.

Very sad considering I got my whole company on Claude Code for them to just ban be like this, with no customer support response.

Post reply on HN