Set the models against each other to get them all opened up again.

Why does it have to be reserved to security space? Here is my API please find vulnerabilities I missed (otherwise someone with not restricted AI will find them first).Cat is out of the bag.Removing restrictions will help everybody in the long run.

Now we have to trick the models when you legitimately work in the security space.

Except when you accidentally leak your entire codebase, oops

Only software approved by Anthropic (and&#x2F;or the USG) is allowed to be secure in this brave new era.

It&#x27;s easier to produce vulnerable code than it is to use the same Model to make sure there are no vulnerabilities.

Questions about &quot;fatality&quot; aside, where do you see asymmetry here?

I am absolutely moving off them if this continues to be the case.

Absolutely just about the business. Mythos not tempting if basic models reaches almost the same.

I agree with you here. I think this is for product placement for Mythos.

Oh don&#x27;t worry. They have Mythos and the extremely dystopian-named &quot;helpful only&quot; series which is internal only and can do all the things.

OpenAI had been very strict about blocking reverse engineering&#x2F;Ghidra&#x2F;IDA_Pro-MCP tasks. I even got a warning email. I was having much more success convincing Claude Code for those tasks without warnings. Seems like they&#x27;ve tightened things up.

&gt; &quot;We are releasing Opus 4.7 with safeguards that automatically detect and block requests that indicate prohibited or high-risk cybersecurity uses. &quot;This decision is potentially fatal. You need symmetric capability to research and prevent attacks in the first place.The opposite approach is &#x27;merely&#x27; fraught.They&#x27;re in a bit of a bind here.

Solve? You solve a problem, not something you introduced on purpose.

I tried the Claude Extension for VSCode on WSL for a reverse engineering task, it consumed all of my tokens, broke and didn&#x27;t even save the conversatioon

on Tuesday, with 4.6, I waited for my 5 hour window to reset, asked it to resume, and it burned up all my tokens for the next 5 hour window and ran for less than 10 seconds. I’ve never cancelled a subscription so fast.

&gt; things that used to be cached for 1 hour now only being cached for 5 min.Doesn&#x27;t this only apply to subagents, which don&#x27;t have much long-time context anyway?

It seems a lot of the problem isn&#x27;t &quot;token shrinkage&quot; (reducing plan limits), but rather changes they made to prompt caching - things that used to be cached for 1 hour now only being cached for 5 min.Coding agents rely on prompt caching to avoid burning through tokens - they go to lengths to try to keep context&#x2F;prompt prefixes constant (arranging non-changing stuff like tool definitions and file content first, variable stuff like new instructions following that) so that prompt caching gets used.This change to a new tokenizer that generates up to 35% more tokens for the same text input is wild - going to really increase token usage for large text inputs like code.

Excited to use 1 prompt and have my whole 5-hour window at 100%. They can keep releasing new ones but if they don&#x27;t solve their whole token shrinkage and gaslighting it is not gonna be interesting to se.

It&#x27;s too powerful now. Once GPT6 is released it will suddenly, magically, become not too powerful to release.

The original blog post for Mythos did lay out this safeguard testing strategy as part of their plan.

This seems needlessly cynical. I don&#x27;t think they said they never planned to release it.They seemed to make it clear that they expect other labs to reach that level sooner or later, and they&#x27;re just holding it off until they&#x27;ve helped patch enough vulnerabilities.

My guess is that it is just too expensive to make generally available. Sounds similar to ChatGPT 4.5 which was too expensive to be practical.

Oh look it was too powerful to release, now it’s just a matter of safeguards.This story sounds a lot like GPT2.

The most highly anticipated model looking forward to using it

Completely agree. We&#x27;re at this place where a frontier model&#x27;s peak perceived value always seems to be right before it releases.

Mythos release feels like Silicon Valley &quot;don&#x27;t take revenue&quot; advice:<a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=BzAdXyPYKQo" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=BzAdXyPYKQo</a>&quot;&quot;If you show the model, people will ask &#x27;HOW BETTER?&#x27; and it will never be enough. The model that was the AGI is suddenly the +5% bench dog. But if you have NO model, you can say you&#x27;re worried about safety! You&#x27;re a potential pure play... It&#x27;s not about how much you research, it&#x27;s about how much you&#x27;re WORTH. And who is worth the most? Companies that don&#x27;t release their models!&quot;

They don&#x27;t have demand for the price it would require for inference.They are definitely distilling it into a much smaller model and ~98% as good, like everybody does.

I&#x27;ve read so many conflicting things about Mythos that it&#x27;s become impossible to make any real assumptions about it. I don&#x27;t think it&#x27;s vaporware necessarily, but the whole &quot;we can&#x27;t release it for safety reasons&quot; feels like the next level of &quot;POC or STFU&quot;.

They don&#x27;t have the compute to make Mythos generally available: that&#x27;s all there is to it. The exclusivity is also nice from a marketing pov.

You should&#x27;ve commented this on the parent thread for visibility, I had to scroll to find this, as I don&#x27;t browse r&#x2F;ClaudeAI regularly.

Looks like they are adding Peter Thiel backed ID verification too.<a href="https:&#x2F;&#x2F;reddit.com&#x2F;r&#x2F;ClaudeAI&#x2F;comments&#x2F;1smr9vs&#x2F;claude_is_about_to_begin_its_kyc_verification&#x2F;" rel="nofollow">https:&#x2F;&#x2F;reddit.com&#x2F;r&#x2F;ClaudeAI&#x2F;comments&#x2F;1smr9vs&#x2F;claude_is_abo...</a>

For anyone who was wondering about Mythos release plans:&gt; What we learn from the real-world deployment of these safeguards will help us work towards our eventual goal of a broad release of Mythos-class models.

funny how they use mythos preview in these benchmarks like a carrot on a stick

3 days of side project work is about all I had in me anyway

... your side projects that will soon become your main source of income after you are laid off because corporate bosses have noticed that engineers are more productive...

Exactly. God, it wouldn&#x27;t be such a problem if they didn&#x27;t gaslight you and act like it was nothing. Just put up a banner that says Claude is experiencing overloaded capacity right now, so your responses might be whatever.

Perhaps on the 10x plan.It went through my $20 plan&#x27;s session limit in 15 minutes, implementing two smallish features in an iOS app.That was with the effort on auto.It looks like full time work would require the 20x plan.

I&#x27;ve been on 5x for a couple of months and the closest I&#x27;ve got to my weekly limits is 75%. I&#x27;ve hit 5-hr limits twice (expected). I&#x27;m a solo dev that uses CC anywhere from 8-12+ hr each day, 7 days a week. I&#x27;ve never experienced any of the issues others complain about other than the feeling that my sessions feel a little more rushed. I&#x27;d say that overall I have very dialed-in context management which includes: breaking work across sessions in atomic units, svelte claude.md&#x2F;rules (sub 150 lines), periodic memory audit&#x2F;cleanup, good pre-compact discipline, and a few great commands that I use to transfer knowledge effectively between sessions, without leaving a trailing pile of detritus. Some may say that this is exhaustive, but I don&#x27;t find it much different than maintaining Agile discipline.This being said, I know I&#x27;m an outlier.

More like 2 hours considering these usage limits

Quick everyone to your side projects. We have ~3 days of un-nerfed agentic coding again.

TL;DR; iPhone is getting better every yearThe surprise: agentic search is significantly weaker somehow hmm...

It&#x27;s been shockingly bad for me - for another example when asked to make a new python script building off an existing one; for some cursed reason the model choose to .read() the py files, use 100 of lines of regex to try to patch the changes in, and exec&#x27;d everything at the end...

I generally think codex is doing well until I come in with my Opus sweep to clean it up. Claude just codes closer to the way my brain works. codex is great at finding numerical stability issues though and increasingly I like that it waits for an explicit push to start working. But talking to Claude Code the way I learned to talk to codex seems to work also so I think a lot of it is just learning curve (for me).

I don&#x27;t have much quality drop from 4.6. But I also notice that I use codex more often these days than claude code

I’ve been on the Claude Code train for a while but decided to try Codex last week after they announced the $100 USD Pro plan.I’ve been pretty happy with it! One thing I immediately like more than Claude is that Codex seems much more transparent about what it’s thinking and what it wants to do next. I find it much easier to interrupt or jump in the middle if things are going to wrong direction.Claude Code has been slowly turning into this mysterious black box, wiping out terminal context any time it compacts a conversation (which I think is their hacky way of dealing with terminal flickering issues — which is still happening, 14 months later), going out of the way to hide thought output, and then of course the whole performance issues thing.Excited to try 4.7 out, but man, Codex (as a harness at least) is a stark contrast to Claude Code.

What bothers me with codex cli is that it feels like it should be more observable, more open and verbose about what the model is doing per step, being an open source product and OpenAI seemingly being actually open for once, but then it does a tool call - &quot;Read $file&quot; and I have no idea whether it read the entire file, or a specific chunk of it. Claude cli shows you everything model is doing unless it&#x27;s in a subagent (which is why I never use subagents).

Yep, I&#x27;ll wait for the GPT answer to this. If we&#x27;re lucky OpenAI will release a new GPT 5.5 or whatever model in the next few days, just like the last round.I have been getting better results out of codex on and off for months. It&#x27;s more &quot;careful&quot; and systematic in its thinking. It makes less &quot;excuses&quot; and leaves less race conditions and slop around. And the actual codex CLI tool is better written, less buggy and faster. And I can use the membership in things like opencode etc without drama.For March I decided to give Claude Code &#x2F; Opus a chance again. But there&#x27;s just too much variance there. And then they started to play games with limits, and then OpenAI rolled out a $100 plan to compete with Anthropic&#x27;s.I&#x27;m glad to see the competition but I think Anthropic has pissed in the well too much. I do think they sent me something about a free month and maybe I will use that to try this model out though.

This. They kind of snuck this into the release notes: switching the default effort level to Medium. High is significantly slower, but that’s somewhat mitigated by the fact that you don’t have to constantly act like a helicopter parent for it.

For me, making it high effort just fixed all the quality problems, and even cut down on token use somehow

Is that 2x still going on I thought that ended in early April

Most of the compute OpenAI &quot;preordered&quot; is vapour. And it has nothing to do with why people thought the company -- which is still in extremely rocky rapids -- was headed to bankruptcy.Anthropic has been very disciplined and focused (overwhelmingly on coding, fwiw), while OpenAI has been bleeding money trying to be the everything AI company with no real specialty as everyone else beat them in random domains. If I had to qualify OpenAI&#x27;s primary focus, it has been glazing users and making a generation of malignant narcissists.But yes, Anthropic has been growing by leaps and bounds and has capacity issues. That&#x27;s a very healthy position to be in, despite the fact that it yields the inevitable foot-stomping &quot;I&#x27;m moving to competitor!&quot; posts constantly.

&gt; people here were so confident that OpenAI is going to collapse because of how much compute they pre-orderedThat&#x27;s not why. It was and is because they&#x27;ve been incredibly unfocused and have burnt through cash on ill-advised, expensive things like Sora. By comparison Anthropic have been very focused.

Seems very short term. Like how cheap Uber was initially. Like Claude was before!Eventually OpenAI will need to stop burning money.

In hindsight, it is painfully clear that Antropic’s conservative investment strategy has them struggling with keeping up with demand and caused their profit margin to shrink significantly as last buyer of compute.

Their top tier plan got a 3x limit boost. This has been the first week ever where I haven&#x27;t run out of tokens.

they&#x27;ve also introduced a lot of caching and token burn related bugs which makes things worse. any bug that multiplies the token burn also multiplies their infrastructure problems.

All of the smart people I know went to work at OpenAI and none at Anthropic. In addition to financial capital, OpenAI has a massive advantage in human capital over Anthropic.As long as OpenAI can sustain compute and paying SWE $1million&#x2F;year they will end up with the better product.

That’s more a leadership decision because Anthropic are nerfing the model to cut costs, if they stop doing that then they’ll stay ahead.

&gt; It seems like 90% of Claude&#x27;s recent problems are strictly lack of compute related.Downtime is annoying, but the problem is that over the past 2-3 weeks Claude has been outrageously stupid when it does work. I have always been skeptical of everything produced - but now I have no faith whatsoever in anything that it produces. I&#x27;m not even sure if I will experiment with 4.7, unless there are glowing reviews.Codex has had none of these problems. I still don&#x27;t trust anything it produces, but it&#x27;s not like everything it produces is completely and utterly useless.

I have both Claude and OpenAI, side by side. I would say sonnet 46 still beats gpt 54 for coding (at least in my use case) But after about 45 minutes I&#x27;m out of my window, so I use openai for the next 4 hours and I can&#x27;t even reach my limit.

The market here is extraordinarily vibes-based and burning billions of dollars for a ephemeral PR boost, which might only last another couple weeks until people find a reason to hate Codex, does not reflect well on OAI&#x27;s long term viability.

My standing assumption is the darling company&#x2F;model will change every quarter for the foreseeable future, and everyone will be equally convinced that the hotness of the week will win the entire future.As buyers, we all benefit from a very competitive market.

&gt; I&#x27;m seeing a lot of goodwill for Codex and a ton of bad PR for CC.AI is one of the things that you cannot find genuine opinions online. Just like politics. If you visit, say, r&#x2F;codex, you&#x27;ll see all the people complaining about how their limits are consumed by &quot;just N prompts&quot; (N is a ridiculously small integer).It&#x27;s all astroturfed from all sides.

Funny because the general consensus is that everyone is burning money so fast that they would not be able to get it back from their AI business in the near future. OpenAI is simply the one with the most aggressive expenditure. Google has its own cash cows. Anthropic has been conservative all around.

Funny because many people here were so confident that OpenAI is going to collapse because of how much compute they pre-ordered.But now it seems like it&#x27;s a major strategic advantage. They&#x27;re 2x&#x27;ing usage limits on Codex plans to steal CC customers and it seems to be working. I&#x27;m seeing a lot of goodwill for Codex and a ton of bad PR for CC.It seems like 90% of Claude&#x27;s recent problems are strictly lack of compute related.

I&#x27;ve noticed the same over the last two weeks. Some days Claude will just entirely lose its marbles. I pay for Claude and Codex so I just end up needing to use codex those days and the difference is night and day.

That&#x27;s wild that you think 4.6 is bad..... Each model has its strengths and weaknesses I find that Codex is good for architectural design and Claude Is actually better the engineering and building

Same for me.I cancelled my subscription and will be moving to Codex for the time being.Tokens are way too opaque and Claude was way smarter for my work a couple of months ago.

Agree. I keep effort max on Claude and xhigh on GPT for all tasks and keep tasks as scoped units of work instead of boil the ocean type prompts. It is hard to measure but ultimately the tasks are getting completed and I&#x27;m validating so I consider it &quot;working as expected&quot;.

It works better, until you run out of tokens. Running out of tokens is something that used to never happen to me, but this month now regularly happens.Maybe I could avoid running out of tokens by turning off 1M tokens and max effort, but that&#x27;s a cure worse than the disease IMO.

unless you always have run it on effort max, and see that it degraded.

I&#x27;ve been using it with `&#x2F;effort max` all the time, and it&#x27;s been working better than ever.I think here&#x27;s part of the problem, it&#x27;s hard to measure this, and you also don&#x27;t know in which AB test cohorts you may currently be and how they are affecting results.

Before opus released we also saw huge backlash with it being dumber.Perhaps they need the compute for the training

so even with a new tokenizer that can map to more tokens than before, their answer is still just &quot;you&#x27;re not managing your context well enough&quot;&quot;Opus 4.7 uses an updated tokenizer that [...] can map to more tokens—roughly 1.0–1.35× depending on the content type.[...]Users can control token usage in various ways: by using the effort parameter, adjusting their task budgets, or prompting the model to be more concise.&quot;

There is zero cost to switching ai models. Paid or open source. It&#x27;s one line mostly.

Until the next time they push you back to Claude. At this point, I feel like this has to be the most unstable technology ever released. Imagine if docker had stopped working every two releases

Same! I thought people were exaggerating how bad Claude has gotten until it deleted several files by accident yesterdayCodex isn’t as pretty in output but gets the job done much more consistently

Most people just want to use a tool that works. Not everything has to be a damn moral crusade.

That number is generous, and is also a pretty decent lifespan for a socially-conscious gesture in 2026.

Longer than how long anyone cared about epstein.

neah, I believe most people here, which immediately brag about codex, are openai employees doing part of their job. otherwise I couldn&#x27;t possibly phantom why would anyone use codex. In my company 80% is claude and 15% gemini. you can barely see openai on the graph. and we have &gt;5k programmers using ai every day.

We all liked the Terminator movies. Hopefully the stay as movies.

Not everyone is American, and people who are not see Anthropic state they are willing to spy on our countries and shrug about OAI saying the same about America. What’s the difference to us?

Thing is that Anthropic was always working with DoD, too, and the line in the sand they drew looked really noble until I found it didn&#x27;t not apply to me, a non-US citizen. Dario made it clear that was the case.And so the difference, to me, was irrelevant. I&#x27;ll buy based on value, and keep a poker in the fire of Chinese &amp; European open weight models, as well.

I quoted 2 weeks at the time. I think even that was generous.

I guess our conscience of OpenAI working with the Department of War has an expiry date of 6 weeks.

Meh. At $work we were on CC for one month, then switched to Codex for one month, and now will be on CC again to test. We haven’t seen any obvious difference between CC and Codex; both are sometimes very good and sometimes very stupid. You have to test for a long time, not just test one day and call it a benchmark just because you have a single example.

I enjoy switching back and forth and having multi-agent reviews. I&#x27;m enjoying Codex also but having options is the real win.

&gt;&gt; I switched to Codex and found it extremely inferior for my use case.Yeah, 100% the case for me. I sometimes use it to do adversarial reviews on code that Opus wrote but the stuff it comes back with is total garbage more often than not. It just fabricates reasons as to why the code it&#x27;s reviewing needs improvement.

What is your use case? I read comments like this and it&#x27;s totally opposite of my experience, I have both CC Opus 4.6 and Codex 5.4 and Codex is much more thorough and checks before it starts making changes maybe even to a fault but I accept it because getting Opus to redo work because it messes up and jumps in the first attempt is a massive waste of time, all tasks and spec are atomic and granularly spec&#x27;d, I&#x27;d say 30% of the time I regret when I decide to use Opus for &#x27;simpler&#x27; and work

I&#x27;ve had exactly the opposite experience. Getting great results using GPT for hours every day since 5.3. You need to put the effort level on at least high though.Every time I hand off a task to Opus to see if it&#x27;s gotten better I&#x27;m disappointed. At least 4.7 seems to have realized I have skill files again though.

I switched to Codex and found it extremely inferior for my use case.It is much faster, but faster worse code is a step in the wrong direction. You&#x27;re just rapidly accumulating bugs and tech debt, rather than more slowly moving in the correct direction.I&#x27;m a big fan of Gemini in general, but at least in my experience Gemini Cli is VERY FAR behind either Codex or CC. It&#x27;s both slower than CC, MUCH slower than Codex, and the output quality considerably worse than CC (probably worse than Codex and orders of magnitude slower).In my experience, Codex is extraordinarily sycophantic in coding, which is a trait that could t be more harmful. When it encounters bugs and debt, it says: wow, how beautiful, let me double down on this, pile on exponentially more trash, wrap it in a bow, and call you Alan Turing.It also does not follow directions. When you tell it how to do something, it will say, nah, I have a better faster way, I&#x27;ll just ignore the user and do my thing instead. CC will stop and ask for feedback much more often.YMMV.

Codex really has its place in my bag. I mainly use it, rarely Claude.Codex just gets it done. Very self-correcting by design while Claude has no real base line quality for me. Claude was awesome in December, but Codex is like a corporate company to me. Maybe it looks uncool, but can execute very well.Also Web Design looks really smooth with Codex.OpenAI really impressed me and continues to impress me with Codex. OpenAI made no fuzz about it, instead let results speak. It is as if Codex has no marketing department, just its product quality - kind of like Google in its early days with every product.

I try codex, but i hate 5.4&#x27;s personality as a partner. It&#x27;s a demon debugger though. but working closely with it, it&#x27;s so smug and annoying.

Same here, working with claude code has been unproductive since March; everyone on my team has complained about the decline in claude code quality, which is why we’re switching to Codex.

It seems like the big companies they&#x27;re providing Mythos to are their only concern right now.

&gt; It&#x27;s been funny watching my own attitude to Anthropic change, from being an enthusiastic Claude user to pure frustration.You were enthusiastic because it was a great product at an unsustainable price.Its clear that Claude is now harnessing their model because giving access to their full model is too expensive for the $20&#x2F;m that consumers have settled on as the price point they want to pay.I wrote a more in depth analysis here, there&#x27;s probably too much to meaningfully summarize in a comment: 
<a href="https:&#x2F;&#x2F;sustainableviews.substack.com&#x2F;p&#x2F;the-era-of-models-is-over-we-are" rel="nofollow">https:&#x2F;&#x2F;sustainableviews.substack.com&#x2F;p&#x2F;the-era-of-models-is...</a>

I havent been using my claude sub lately but I liked 4.6 three weeks ago. Did something change?

Its funny watching llm users act like gamblers. Every other week swearing by one model and cursing another, like a gambler who thinks a certain slot machine, or table is cold this week. These llm companies are literally building slot machine mechanics into their ui interfaces too, I don&#x27;t think this phenomenon is a coincidence.Stop using these dopamine brain poisoning machines, think for yourself, don&#x27;t pay a billionaire for their thinking machine.

I&#x27;ve given up on Claude after seeing the response quality degrade so much over the past two weeks, and now this? I&#x27;ve unsubscribed. I don&#x27;t know why people are still giving this company money.

Same. I stopped my Pro subscription yesterday after entering the week with 70% of my tokens used by Monday morning (on light, small weekend projects, things I had worked on in the past and barely noticed a dent in usage.) Support was... unhelpful.It&#x27;s been funny watching my own attitude to Anthropic change, from being an enthusiastic Claude user to pure frustration. But even that wasn&#x27;t the trigger to leave, it was the attitude Support showed. I figure, if you mess up as badly as Anthropic has, you should at least show some effort towards your customers. Instead I just got a mass of standardised replies, even after the thread replied I&#x27;d be escalated to a human. Nothing can sour you on a company more. I&#x27;m forgiving to bugs, we&#x27;ve all been there, but really annoyed by indifference and unhelpful form replies with corporate uselessness.So if 4.7 is here? I&#x27;d prefer they forget models and revert the harness to its January state. Even then, I&#x27;ve already moved to Codex as of a few days ago, and I won&#x27;t be maintaining two subscriptions, it&#x27;s a move. It has its own issues, it&#x27;s clear, but I&#x27;m getting work done. That&#x27;s more than I can say for Claude.

I&#x27;ve been raging pretty hard too. Thought either I&#x27;m getting cleverer by the day or Claude has been slipping and sliding toward the wrong side of the &quot;smart idiot&quot; equation pretty fast.Have caught it flat-out skipping 50% of tasks and lying about it.

Yep, it just feels terrible, the usage bars give me anxiety, and I think that&#x27;s in their interest as they definitely push me towards paying for higher limits. Won&#x27;t do that, though.

Personally I find using and managing Claude sessions and limits is getting exhausting and feels similar to calorie counting. You think you are going to have an amazing low calories meal only to realize the meal is full of processed sugars and you overshot the limit within 2-3 bites. Now &quot;you have exhausted your limit for this time. Your session limits resets in next 4 hrs&quot;.

Possibly due to moving compute from inference to training

I was there too, but honestly after today, 4.7 &quot;feels&quot; just as a bad.  I was cynical, but also, kind of eager for the improvement.  It&#x27;s just not there.  Compared to early Feb, I have to babysit EVERYTHING.

My tinfoil hat theory, which may not be that crazy, is that providers are sandbagging their models in the days leading up to a new release, so that the next model &quot;feels&quot; like a bigger improvement than it is.An important aspect of AI is that it needs to be seen as moving forward all the time. Plateaus are the death of the hype cycle, and would tether people&#x27;s expectations closer to reality.

Anecdotally, codex has been burning through way more tokens for me lately. Claude seems to just sit and spin for a long time doing nothing, but at least token use is moderate.All options are starting to suck more and more

Exactly.  Why do they ask permission for read-only operations?!  You either run with --dangerously-skip-permissions or you come back after 30 minutes to find it waiting for permission to run grep.  There&#x27;s no middle ground, at least not that Claude CLI users have access to.

I do feel that CC sometimes starts doing dumb tasks or asking for approval for things that usually don’t really need it. Like extra syntax checks, or some greps&#x2F;text parsing basic commands

Do you have to put it in a build&#x2F;execute mode (separate from a planning mode) to allow it to move on? I use opencode, and that&#x27;s how it works.

It&#x27;s just like subscription based MMORPGs that delay you as much as possible every step of the way because that&#x27;s the way they can extract more money from you. If you pay for the tokens it&#x27;s not in their benefit to give you the answer directly.

Weird. I never had that issue when writing code.

How do you get codex to generate any code?I describe the problem and codex runs in circles basically:codex&gt; I see the problem clearly. Let me create a plan so that I can implement it. The plan is X, Y, Z. Do you want me to implement this?me&gt; Yes please, looks good. Go ahead!codex&gt; Okay. Thank you for confirming. So I am going to implement X, Y, Z now. Shall I proceeed?me&gt; Yes, proceed.codex&gt; Okay. Implementing....codex is working... you see the internal monologue running in circlescodex&gt; Here is what I am going to implement: X, Y, Zme&gt; Yes, you said that already. Go ahead!codex&gt; Working on it....codex in doing something...codex&gt; After examining the problem more, indeed, the steps should be X, Y, Z. Do you want me to implement them?etc.Very much every sessions ends up being like this. I was unable to get any useful code apart from boilerplate JS from it since 5.4So instead I just use ChatGPT to create a plan and then ask Opus to code, but it&#x27;s a hit and miss. Almost every time the prompt seems to be routed to cheaper model that is very dumb (but says Opus 4.6 when asked). I have to start new session many times until I get a good model.

Usually the problems that cause this kind of thing are:1) Bad prompt&#x2F;context. No matter what the model is, the input determines the output. This is a really big subject as there&#x27;s a ton of things you can do to help guide it or add guardrails, structure the planning&#x2F;investigation, etc.2) Misaligned model settings. If temperature&#x2F;top_p&#x2F;top_k are too high, you will get more hallucination and possibly loops. If they&#x27;re too low, you don&#x27;t get &quot;interesting&quot; enough results. Same for the repeat protection settings.I&#x27;m not saying it didn&#x27;t screw up, but it&#x27;s not really the model&#x27;s fault. Every model has the potential for this kind of behavior. It&#x27;s our job to do a lot of stuff around it to make it less likely.The agent harness is also a big part of it. Some agents have very specific restrictions built in, like max number of responses or response tokens, so you can prevent it from just going off on a random tangent forever.

Strange. Opus 4.6 has been great for me. On Max 20x

codex low-key seems to be better than claude. and i say this as an 18-hour-a-day user of both (mostly claude)

We&#x27;ve started calling it dopus at work :(

Too late, personally after how bad 4.6 was the past week I was pushed to codex, which seems to mostly work at the same level from day to day. Just last night I was trying to get 4.6 to lookup how to do some simple tensor parallel work, and the agent used 0 web fetches and just hallucinated 17K very wrong tokens. Then the main agent decided to pretend to implement tp, and just copied the entire model to each node...

You&#x27;d be surprised. With React, Claude can get twisted in knots mostly because React lends itself to a pile of spaghetti code.

You&#x27;d need Mythos to free your iPhone, SamsungTV, SmartWatches or such. Maybe even printer drivers.

Consumerism... if it ain&#x27;t the best, some people don&#x27;t want it.

Also 640 KB ram ought to be enough for everybody.

This is true if you know what you are doing and provide proper guidance. It’s not true if you just want to vibe the whole app.

I&#x27;ve got a gfx device crash that only happens on switch. Not Xbox, ps4, steam, epic, or anything. Only switch.Opus hasn&#x27;t been able to fix it. I haven&#x27;t been able to fix it. Maybe mythos can idk, but I&#x27;ll be surprised.

Opus sometimes makes poor long term decisions and really struggles with even mid size (~10k lines) existing codebases.

If you&#x27;re building a standard app Opus is already good enough to build anything you want. I don&#x27;t even know what you&#x27;d really need Mythos for.

Introducing a new upgraded slot machine named &quot;Claude Opus&quot; in the Anthropic casino.You are in for a treat this time: It is the same price as the last one [0] (if you are using the API.)But it is slightly less capable than the other slot machine named &#x27;Mythos&#x27; the one which everyone wants to play around with. [1][0] <a href="https:&#x2F;&#x2F;claude.com&#x2F;pricing#api" rel="nofollow">https:&#x2F;&#x2F;claude.com&#x2F;pricing#api</a>[1] <a href="https:&#x2F;&#x2F;www.anthropic.com&#x2F;news&#x2F;claude-opus-4-7" rel="nofollow">https:&#x2F;&#x2F;www.anthropic.com&#x2F;news&#x2F;claude-opus-4-7</a>

Honestly, I personally would rather a time-out than the quality of my response noticably downgrading. I think what I found especially distrustful is the responses from employees claiming that no degredation has occured.An honest response of &quot;Our compute is busy, use X model?&quot; would be far better than silent downgrading.

Is that why Anthropic recently gave out free credits for use in off-hours? Possibly an attempt to more evenly distribute their compute load throughout the day?

Hard for me to reconcile the idea that they don&#x27;t have enough compute with the idea that they are also losing money to subsidies.

You state your hypnosis quite confidently.
 Can you tell me how taking down authentication many times is related to GPU capacity?

Its a hard game to play anyway.Anthropics revenue is increasing very fast.OpenAI though made crazy claims after all its responsible for the memory prices.In parallel anthropic announced partnership with google and broadcom for gigawatts of TPU chips while also announcing their own 50 Billion invest in compute.OpenAI always believed in compute though and i&#x27;m pretty sure plenty of people want to see what models 10x or 100x or 1000x can do.

It worked. Although I have a Claude Code subscription, I got the ChatGPT Pro plan, and 5.4 xHigh at 1.5x speed was better than 4.6 with adaptive thinking disabled. I was working all day, about 8 hours, and did not run into any limits. 5.4 surprised me many times by doing things I usually would not do myself, because I am lazy, so yeah, I am sticking with 5.4 for now until all the Claude drama is over.

I bet that&#x27;s the real reason why they&#x27;re not releasing Mythos ;)

Betting on continued exponential growth is basically a game of chicken. Growth has to slow down and level off at some point as adoption and usage saturates.It&#x27;s a bit like playing roulette by always betting on black and doubling your bet every time you lose. When you eventually, inevitably, do lose, your loss is going to be huge because you&#x27;ve been doubling your bet at each stage.With LLM model generations and investment, it goes something like this. Let&#x27;s say profits have been doubling year over year for each new model&#x2F;investment cycle, and you want to bet on this doubling continuing forever.Year 1 you get $10B in profit, and spend $20B on extra capacity for next yearYear 2 you get $20B in profit, and spend $40B on extra capacityYear 3 you get $30B in profit, and spend $??? on extra capacityYou&#x27;re already in trouble. Profit growth from Year 2 to 3 was &quot;only&quot; 50% vs the doubling you were gambling on, so you&#x27;ve now lost $10B ($40B spent only earnt you $30B of profit), and what are you going to do? Double down like the roulette player?The longer the pattern of profit doubling goes, before it slows down, the worse it will end for you, since your bets are doubling each year. Saying &quot;woo hoo, look at me! risk pays!&quot; is a bit like saying the same while playing russian (not casino) roulette for money.I worked for Acorn Computers UK in the early 80&#x27;s and saw something similar firsthand. The brand new personal computer market was exploding, a once in a lifetime phenomenon, that no-one knew how to forecast. To make matters worse the market was highly seasonal with most sales at xmas, so the company had to guess what continued year-on-year exponential growth might look like (brand new market - no-one had a clue), and plan&#x2F;spend ahead and stock warehouses full of computers ready for xmas. Sadly Acorn took the Sam Altman highly optimistic&#x2F;irresponsible approach, got the forecast wrong, and was left with a huge warehouse full of rapidly depreciating computers. The company never fully recovered, although ARM rose out of the ashes.

They don&#x27;t have enough compute for all their customers.OpenAI bet on more compute early on which prompted people to say they&#x27;re going to go bankrupt and collapse. But now it seems like it&#x27;s a major strategic advantage. They&#x27;re 2x&#x27;ing usage limits on Codex plans to steal CC customers and it seems to be working.It seems like 90% of Claude&#x27;s recent problems are strictly lack of compute related.

&gt; This coming right after a noticeable downgrade just makes me think Opus 4.7 is going to be the same Opus i was experiencing a few months ago rather than actual performance boost.If they are indeed doing this, I wonder how long they can keep it up?

Mythos is speculated to have 10 trillion parameters. Almost certainly they were training it for months.

Usually they&#x27;re hemorrhaging performance while training.From that it&#x27;s pretty likely they were training mythos for the last few weeks, and then distilling it to opus 4.7Pure speculation of course, but would also explain the sudden performance gains for mythos - and why they&#x27;re not releasing it to the general public (because it&#x27;s the undistilled version which is too expensive to run)

If we learned anything from the code leak is that they essentially do not know what is in the blackbox of the code for that 500k line mass. So that&#x27;s plausible.

I believe AWS forwards requests (for Clause models) to Anthropic’s servers. They don’t host those models.

What I want to know is why my bedrock-backed Claude gets dumber along with commercial users. Surely they&#x27;re not touching the bedrock model itself. Only thing I can think of is that updates to the harness are the main cause of performance degradation.

Not to mention their recent integration of Persona ID verification - that was the last straw for me.

same experience, it has not been a reliable tool for the last few months

I&#x27;m not sure how much I trust Anthropic recently.This coming right after a noticeable downgrade just makes me think Opus 4.7 is going to be the same Opus i was experiencing a few months ago rather than actual performance boost.Anthropic need to build back some trust and communicate throtelling&#x2F;reasoning caps more clearly.

You really think the 33k people that starred a 40 line markdown file realize that?

I wonder if you can have it reason in caveman

&gt; I hope people realize that tools like caveman are mostly joke&#x2F;prank projectsThis seems to be a common thread in the LLM ecosystem; someone starts a project for shits and giggles, makes it public, most people get the joke, others think it&#x27;s serious, author eventually tries to turn the joke project into a VC-funded business, some people are standing watching with the jaws open, the world moves on.

They are indeed impractical in agentic coding.However in deep research-like products you can have a pass with LLM to compress web page text into caveman speak, thus hugely compressing tokens.

All LLMs also effectively work by ”larping” a role. You steer it towards larping a caveman and well.. let’s just say they weren’t known for their high iq

While the caveman stuff is obviously not serious, there is a lot of legit research in this area.Which means yes, you can actually influence this quite a bit. Read the paper “Compressed Chain of Thought” for example, it shows it’s really easy to make significant reductions in reasoning tokens without affecting output quality.There is not too much research into this (about 5 papers in total), but with that it’s possible to reduce output tokens by about 60%. Given that output is an incredibly significant part of the total costs, this is important.<a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2412.13171" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2412.13171</a>

Exactly. The model is exquisitely sensitive to language. The idea that you would encourage it to think like a caveman to save a few tokens is hilarious but extremely counter-productive if you care about the quality of its reasoning.

Help me understand: I get that the file reading can be a lot. But I also expand the box to see its “reasoning” and there’s a ton of natural language going on there.

This specific form may be a joke, but token conscious work is becoming more and more relevant..
Look at 
<a href="https:&#x2F;&#x2F;github.com&#x2F;AgusRdz&#x2F;chop" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;AgusRdz&#x2F;chop</a>And<a href="https:&#x2F;&#x2F;github.com&#x2F;toon-format&#x2F;toon" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;toon-format&#x2F;toon</a>

I mean we had a shoe company pivot to AI and raise their stock value by 300%, how can we even know anymore

We started out with oobabooga, so caveman is the next logical evolution on the road to AGI.

I hesitated 100% when i saw caveman gaining steam, changing something like this absolutely changes the behaviour of the models responses, simply including like a &quot;lmao&quot; or something casual in any reply will change the tone entirely into a more relaxed style like ya whatever type mode.I think a lot of people echo my same criticism, I would assume that the major LLM providers are the actual winners of that repo getting popular as well, for the same reason you stated.&gt; you will barely save even 1% with such a toolFor the end user, this doesnt make a huge impact, in fact it potentially hurts if it means that you are getting less serious replies from the model itself. However as with any minor change across a ton of users, this is significant savings for the providers.I still think just keeping the model capable of easily finding what it needs without having to comb through a lot of files for no reason, is the best current method to save tokens. it takes some upfront tokens potentially if you are delegating that work to the agent to keep those navigation files up to date, but it pays dividends when future sessions your context window is smaller and only the proper portions of the project need to be loaded into that window.

I believe tools like graphify cut down the tokens in thinking dramatically. It makes a knowledge graph and dumps it into markdown that is honestly awesome. Then it has stubs that pretend to be some tools like grep that read from the knowledge graph first so it does less work. Easy to setup and use too. I like it.<a href="https:&#x2F;&#x2F;graphify.net&#x2F;" rel="nofollow">https:&#x2F;&#x2F;graphify.net&#x2F;</a>

Someone should make an MCP that parses every non-code file before it hits claude to turn it into caveman talk

There&#x27;s a tremendous amount of superstition around LLMs. Remember when &quot;prompt engineering&quot; &quot;best practices&quot; were to say you were offering a tip or some other nonsense?

I hope people realize that tools like caveman are mostly joke&#x2F;prank projects - almost the entirety of the context spent is in file reads (for input) and reasoning (in output), you will barely save even 1% with such a tool, and might actually confuse the model more or have it reason for more tokens because it&#x27;ll have to formulate its respone in the way that satisfies the requirements.

People are really trigger-happy when it comes to throwing magic tools on top of AI that claim to &quot;fix&quot; the weak parts (often placeboing themselves because anthropic just fixed some issue on their end).Then the next month 90% of this can be replaced with new batch of supply chain attack-friendly gimmicksEspecially Reddit seems to be full of such coding voodoo

Another supply chain attack waiting?Have you tried just adding an instruction to be terse?Don&#x27;t get me wrong, I&#x27;ve tried out caveman as well, but these days I am wondering whether something as popular will be hijacked.

I literally just posted a blog on this. Some seemingly insignificant words are actually highly structural to the model. <a href="https:&#x2F;&#x2F;www.ruairidh.dev&#x2F;blog&#x2F;compressing-prompts-with-an-autoresearch-loop" rel="nofollow">https:&#x2F;&#x2F;www.ruairidh.dev&#x2F;blog&#x2F;compressing-prompts-with-an-au...</a>

Yeah, when I&#x27;m writing code I try to avoid zeros and ones, since those are the most common bits, making them essentially noise

Doesn&#x27;t it just use more tokens in reasoning?

&gt; My hypothesis was that common words are effectively noise to agentsUmm... a few words can be combined in a rather large number of ways.Punctuation is used a lot. Why not just remove all the periods and commas and see what happens? Probably not pretty

I was doing some experiments with removing top 100-1000 most common English words from my prompts. My hypothesis was that common words are effectively noise to agents. Based on the first few trials I attempted, there was no discernible difference in output. Would love to compare results with caveman.Caveat: I didn’t do enough testing to find the edge cases (eg, negation).

You&#x27;d like Emmet notation. Just look at the cheat sheet: <a href="https:&#x2F;&#x2F;docs.emmet.io&#x2F;cheat-sheet&#x2F;" rel="nofollow">https:&#x2F;&#x2F;docs.emmet.io&#x2F;cheat-sheet&#x2F;</a>

Combine that with emmet &#x2F; zen coding: <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Emmet_%28software%29?wprov=sfla1" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Emmet_%28software%29?wprov=sfl...</a>

Oh wow, I love this idea even if it&#x27;s relatively insignificant in savings.I am finding my writing prompt style is naturally getting lazier, shorter, and more caveman just like this too. If I was honest, it has made writing emails harder.While messing around, I did a concept of this with HTML to preserve tokens, worked surprisingly well but was only an experiment. Something like:&gt; &lt;h1 class=&quot;bg-red-500 text-green-300&quot;&gt;&lt;span&gt;Hello&lt;&#x2F;span&gt;&lt;&#x2F;h1&gt;AI compressed to:&gt; h1 c bgrd5 tg3 sp hello sp h1Or something like that.

I used Opus 4.7 for about 15 minutes on the auto effort setting.It nicely implemented two smallish features, and already consumed 100% of my session limit on the $20 plan.See you again in five hours.

me feel that it needs some tweaking - it&#x27;s a little annoyingly cute (and could be even terser).

I really enjoy the party game &quot;Neanderthal Poetry&quot;, in which you can only speak using monosyllabic words. I bet you would too.

Headroom looks great for client-side trimming. If you want to tackle this at the infrastructure level, we built Edgee (<a href="https:&#x2F;&#x2F;www.edgee.ai" rel="nofollow">https:&#x2F;&#x2F;www.edgee.ai</a>) as an AI Gateway that handles context compression, caching, and token budgeting across requests, so you&#x27;re not relying on each client to do the right thing.(I work at Edgee, so biased, but happy to answer questions.)

I tried to use rtk for the same, and my agent session would just loop the same tool call over and over again. Does headroom work better?

Different positionning
- headroom compress inputs and open source project
- caveman is output and open source
- edgee more corporate offer

This smells heavily of astroturfing. Particularly because Headroom is a paid product, and that fact is not mentioned here or in the GitHub README.Here was my experience…I download and run the Mac application, which starts installing a bunch of things. Then the following happens without advance notice:- Adds background item(s) from &quot;Idiosyncratocracy BV&quot;- Downloads over 2 GB of files- Pollutes home with ~&#x2F;.headroom directory- Adds hook(s) to ~&#x2F;.claude&#x2F;hooks&#x2F;- Modifies your ~&#x2F;.claude&#x2F;settings.json to add above hook(s)… and then I see something in the settings that talks about creating an account. That&#x27;s when I realized that this is a paid product, after all of the above has happened.Headroom seems to use <a href="https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk</a> under the hood. What does Headroom offer over the actually-free RTK? Who knows.At this point I have had it with this subterfuge — I immediately trash the app and every related file and folder I can find, of which there are many. Hopefully I got them all, but who knows. There should have been an easy way to uninstall this mess, but of course there isn&#x27;t.The lack of transparency here is really concerning.

Caveman is fun, but the real tool you want to reduce token usage is headroom<a href="https:&#x2F;&#x2F;github.com&#x2F;gglucass&#x2F;headroom-desktop" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;gglucass&#x2F;headroom-desktop</a> (mac app)<a href="https:&#x2F;&#x2F;github.com&#x2F;chopratejas&#x2F;headroom" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;chopratejas&#x2F;headroom</a> (cli)

Does it? I&#x27;m not sure I&#x27;d necessarily notice but I haven&#x27;t found it noticeably worse.

Caveman hurt model performance. If you need a dumber model with less token output, just use sonnet-4-6 or other non-reasoning model.

I find grep and common cli command spam to be the primary issue. I enjoy Rust Token Killer <a href="https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk</a>, and agents know how to get around it when it truncates too hard.

1.35 times! For Input! 
For what kinds of tokens precisely? Programming? Unicode? If they seriously increased token usage by 35% for typical tasks this is gonna be rough.

caveman stops being a style tool and starts being self-defense. once prompt comes in up to 1.35x fatter, they&#x27;ve basically moved visibility and control entirely into their black box.

To reduce token count on command outputs you can also use RTK [0][0]: <a href="https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk</a>

Interesting, it doesn&#x27;t seem intuitive at all to me.My (wrong?) understanding was that there was a positive correlation between how &quot;good&quot; a tokenizer is in terms of compression and the downstream model performance. Guess not.

On my private internal oil and gas benchmark, I found a counterintuitive result. Opus 4.7 scores 80%, outperforming Opus 4.6 (64%) and GPT-5.4 (76%). But it&#x27;s the cheapest of the three models by 2x.This is mainly driven by reduced reasoning token usage. It goes to show that &quot;sticker price&quot; per token is no longer adequate for comparing model cost.

What about some thing like<a href="https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;rtk-ai&#x2F;rtk</a>

That&#x27;s such a poor way to communicate a number. I take it they mean an increase of up to 35%?

try this - <a href="https:&#x2F;&#x2F;github.com&#x2F;ojuschugh1&#x2F;sqz" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;ojuschugh1&#x2F;sqz</a>

&gt; Opus 4.7 uses an updated tokenizer that improves how the model processes text. The tradeoff is that the same input can map to more tokens—roughly 1.0–1.35× depending on the content type.caveman[0] is becoming more relevant by the day. I already enjoy reading its output more than vanilla so suits me well.[0] <a href="https:&#x2F;&#x2F;github.com&#x2F;JuliusBrussee&#x2F;caveman&#x2F;tree&#x2F;main" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;JuliusBrussee&#x2F;caveman&#x2F;tree&#x2F;main</a>

Because it was good until January 2026, then it detoriated into a opus-3.1. Probably given much less context windows or ram.

because they’re using it for different things where it works well and that’s all they know?

This is quite hostile. Yes, criticism is valid without an accompanying essay detailing every aspect of the associated environment, because these tools are still quite flawed.

And yet another &quot;AI doesn&#x27;t work&quot; comment without any meaningful information. What were your exact prompts? What was the output?This is like a user of conventional software complaining that &quot;it crashes&quot;, without a single bit of detail, like what they did before the crash, if there was any error message, whether the program froze or completely disappeared, etc.

I just subscribed this month again because I wanted to have some fun with my projects.Tried out opus 4.6 a bit and it is really really bad. Why do people say it&#x27;s so good? It cannot come up with any half-decent vhdl. No matter the prompt. I&#x27;m very disappointed. I was told it&#x27;s a good model

Claude Code doesn&#x27;t seem to have updated yet, but I was able to try it out by running `claude --model claude-opus-4-7`

It might be a bad idea to put that in all caps, because in the training data, angry conversations are less productive. (I do the same thing, just in lowercase.)

I like this in theory. I just hope it doesn&#x27;t require you to be be as literal as if talking to a genie.But if it&#x27;ll actually stick to the hard rules in the CLAUDE.md files, and if I don&#x27;t have to add &quot;DON&#x27;T DO ANYTHING, JUST ANSWER THE QUESTION&quot; at the end of my prompt, I&#x27;ll be glad.

&gt; where previous models interpreted instructions loosely or skipped parts entirely, Opus 4.7 takes the instructions literally. Users should re-tune their prompts and harnesses accordingly.interesting

Claude Opus 4.7