I think the difference between the Anthropic token maximization approach (vibe code all the things!) and OpenAI's focus on efficiency, terseness and token reduction are going to be the defining features of who wins the long-term race.
My money is on the more efficient solution. Even if Anthropic can win some benchmarks by using 3x tokens over 3x time, it is a terrible base to build toward the future. Users are no longer willing to wait exponentially long for linear improvements. And as we see from some of the Chinese models, they can quickly distill frontier models with the tax of being slower and more token-guzzling, while retaining most of the quality. The real differentiators are becoming speed and efficiency, which translate to cost and user velocity more than incremental capability improvements.
It's great that frontier models can solve complex math equations, but bread and butter LLM usage (where the money is made) has already shifted from "I need the best always" to "what solves my day-to-day problems quickly and consistently". Fable usage as a percentage is flat-lining. We are already at the point where output quality is negligible. What wins going forward is cost, speed, consistency and the compounding effects of "softer" improvements to the harness.
Not to sound like a mark, I try to not get attached to any of these providers.
I've jumped between copilot, claude, gemini and chatgpt since the start of the year. chatgpt wasn't even worth looking at early this year.
Anthropic has the smarter models for sure, and seems to be default in corporate. However, the amount of budget you get with GPT as a user is much better, the harness feels more polished, and the models are faster. They are also much nicer to work with, I can just read the output for the most part. With claude I get pages of text and need to skim to find where the actual information i need to care about lies. So much more cognitive overhead.
Sol is smart enough for anything I've thrown at it, it's not one-shotting like fable, but I'm more willing to actually go back and forth with it, and it's likely producing better output to keep a human in the loop rather than trying to solve the world independently and making multiple incorrect assumptions.
I think GPT sees the market changing and is correctly repositioning themselves. Anthropic is down the wrong road, and if they don't correct course quickly I'm sure many of those enterprise contracts will start pivoting.
I really would like to see OpenAI’s focus on efficiency but everytime I use Codex, it wastes tokens like there’s no tomorrow, hitting week limit in a day, where I’m able to use Claude just fine. Maybe it’s based on the codebase, I don’t know, but I have better results with Claude than Codex.
My boss uses opus and gets good results when I use it always burns tokens. The other way is also true too I get great results with sol/terra but my boss does not.
I had the opposite experience. Claude models and its harness feel like they are set to eat tokens for everything, especially if it’s ultracode effort. I have seen it spawn 6 agents and eat my 4-hour quota right in front of me. Codex is on point and follows instructions well even with max effort. I like my analogy of Claude being garrulous and Codex being laconic.
If I see more limit hits in same sitting session, I have to rearrange my workflows. my experiments with qwen and deepseek have been good, cant wait to try glm and other models.
Just before I canceled my 20x Max sub I had a day where I had ~15% of my weekly I was trying to burn. I set it on ultracode and because I had set the max agents 32 for another project and forgot, the five research agents ended up spinning up a total of 26 subagents and burned through the remainder of my weekly in the span of 20 minutes before I noticed and shut it down.
You've articulated what I found unsettling about Boris' pov that "coding is a solved problem". I listened to a few of his talks and was instinctively off-put by that sentiment. I figured a fellow programmer would understand and speak on the nuances.
Granted it did make me think about my biases and to lean into more future facing inevitabilities. But you've nailed it, for Boris and Anthropic, they are betting that coding is a solved problem in the sense that any person can one-shot any random idea and the output will be in some abstract sense "good". And then at what cost and toward what end?
On the one hand it feels true. On the other I ask - what good software have anthropic, or anyone else, produced that was fully vibed?
As someone who does near 100% of my coding via LLM these days, i still find that for anything complex i am still looking at and thinking in terms of code. Im still quality checking and steering at some interval via code. And im still not sure how or whether i can replicate that level of thought without still dealing in code at times.
It's a scale thing. with one fairly simple Android app, it's doable. anything Enterprisey is a nightmare. maybe that's the lesson and the problem is not with the llms but with accidental complexity. Just right now I feel like I'm one mythical LLM minute away from the next clean pull request...
In this last week I've started reviewing code more and in just a short time, I've found 3 fairly simple things that were introduced by Claude that were not wrong per se, however, they were very inefficient and didn't address the root of the issue. I still think we're in a place where the output looks good as long as you don't look under the hood or keep it scoped to small, vibe-coded projects. Once you get beyond that it can fall apart. Anthropic must have a large codebase by now though. Yet I haven't seen much released from them about how they actually work day to day on development.
What makes you think Antropic has a large code base? what do they do exactly that would make them need a large code base? Or maybe the word Large can be understood in different way?
it'll all be centralized models justifying their capability against local inference in, maybe 3 years or less
once we have a bit more memory fab capacity and the insane bottomless investment in AI giants realizes there is a bottom, local hardware will catch up with model performance to the extent that centralized inference will be downgraded to special cases or for orgs that find it cheaper than buying expensive hardware
but really most power users are going to have 1 TB unified RAM and local models that will do well enough
for light office use you can still have a cheap laptop and a claude subscription
Seems like it wouldn't be hard for Anthropic to tweak a few prompts or RL pipelines to tune for terseness and token-efficiency if that stays as something that consumers want.
I find it unlikely that there's some fundamental property of OpenAI's models' "personality" or style which Anthropic (or any other serious AI firm) wouldn't be able to match if they wanted to.
Claude used to be 10x more efficient.
I think there are no barriers of entry between one coding agent and the other.
Therefore, I expect them to reverse as soon as people switch.
At least this is what I will do.
There is no evidence regarding distillation. It is impossible to distill a model in just a month which was the gap between fable and Kimi k3. Anthropic wouldn't even keep up with the load. It is just another example of American exceptionalism.
“In one notable technique, their prompts asked Claude to imagine and articulate the internal reasoning behind a completed response and write it out step by step—effectively generating chain-of-thought training data at scale. We also observed tasks in which Claude was used to generate censorship-safe alternatives to politically sensitive queries like questions about dissidents, party leaders, or authoritarianism, likely in order to train DeepSeek’s own models to steer conversations away from censored topics. By examining request metadata, we were able to trace these accounts to specific researchers at the lab.”
—
“You are an expert data analyst combining statistical rigor with deep domain knowledge. Your goal is to deliver data-driven insights — not summaries or visualizations — grounded in real data and supported by complete and transparent reasoning.” (variations appearing 10s of 1000s of times)
Does Dario have the same relationship with the truth as Sam? (Their companies pirate books and develop products that compete at some level with those books’ authors, so obviously neither are that wonderfully trustworthy, so maybe “no evidence” meant you don’t believe this evidence rather than you weren’t aware of it. I would understand and respect your lack of belief!)
They are not paying billions anymore. They just switched to buying and then destroying books. Arguably that's worse than training off of Anna's archive which Chinese companies still do. This comment is another example of American exceptionalism.
That's what any harness could do. By itself it's not evidence of distillation. Just mentioning Deepseek in there is just fear mongering, again American exceptionalism, which creates the belief only the US has the right to own cutting edge technology. It's a disservice to the entire world, and china happens to be the only challenger to that belief. They surely also asked for tasks to avoid mentioning Trump's involvement in the Epstein files, but don't mention those, due to American exceptionalism. Dario has stated that he is an American exceptionalist. He powers Israeli/US bombings in Gaza, but only cares when China does the same.
This is evidence only if you're an American exceptionalist.
Oh no, I use the top shelf models every day and I really think they have a lot of room to improve in pretty much every regard. I suspect Fable usage is flatlining because it's not that good comparitively and way too expensive.
> My money is on the more efficient solution. Even if Anthropic can win some benchmarks by using 3x tokens over 3x time, it is a terrible base to build toward the future. Users are no longer willing to wait exponentially long for linear improvements.
As much as I wish you were right, everything about software economics for the last 30+ years has favored _less efficient software_. Traditional hardware has been optimized for traditional software for decades and we still see bloated software win consistently. LLM hardware is at the start of its cycle, with abundant low-hanging fruit to conquer--I would expect the pro-bloat dynamics to weigh even more heavily in the LLM space than in the traditional software space.
It's hard to put a number on it, but even accounting for all the time in meetings, talking with stakeholders and developers etc I'm over 10% more productive overall. I earn substantially more than $2000/month, the ROI is there
It's only expensive compared to the currently very strong offering from OpenAI. Or other models - my hobby projects are all on DeepSeek
What you’re saying is that they should be charging $2000 or more per month then.
Perhaps that’s true, but it’s a price tag that’s a lot harder to swallow for many orgs than $200/mo, and would require some hard justification for how your increased productivity contributes to the business bottom line.
I’ll agree that you can probably do that with hard numbers. I am skeptical that most $200/mo users could.
I believe it's been crickets because open-source/DIY solutions almost never become mainstream. Devs tend to think everyone wants a self-service to save a few bucks.
Most ordinary people don't care that much about privacy and if there is a solution that just works well enough to cover their needs and costs just fair enough to be affordable, they are going to use that. I feel like this is why fast food is such a thing. There are far better food options out there that either require a little bit more effort or money. Your post on Linkedin is akin to someone claiming in 1960's America that home-cooked burgers are the future, despite McDonalds gaining ground.
It is also why crypto wallets and the like never took off, people don't want to give up the convenience of a bank account, understandably.
I switched when 4.7 dropped, after over a year on the service. It was an immediate improvement - not even 1 hour before Codex was giving me better results with less effort.
They're absolutely best in a chocolate-in-my-peanutbutter situation - I use them to adversarially review each other's changes and ping pong the branch back and forth - A makes changes, B reviews, B makes changes, A reviews. Together they're much smarter than either individually - anthropic models tend towards more creativity, openai models can get bogged down in the nitty gritty.
Stick with xhigh rather than turbo for most tasks though, now that the Age of Abundant Resets seems to be well and truly over for good (and 5h limits are supposedly returning soon).
I haven’t had any issues with Fable or Opus code-wise, but the way it has started writing recently has almost become incomprehensible.
At the end of a long session it will start saying stuff like “There are smoke tests on the foundation-gates that are left for the cutting seam checks on these domains, which is genuinely your decision”
Never before the past couple months have I ever not been able to understand wtf it’s even saying lol
I was working on a project recently that required the attribution of a data source, and it added to the "licenses" page of the project something like "We use <blah> and per their terms we owe an acknowledgement of attribution to you, the user."
So in addition to the stuff it says in a session there's gobbledy gook that it prints out in copy as well.
For anyone on the fence, I was a hardcore CC user since it was released. I made the full switch to 5.6 sol and Codex about a month ago.
It's the better experience. The limits are way higher (I almost never burn through my $200/m plan), and the output is better than Opus 4.8 (Opus 5 is completely unusable for me).
Also try a hundred bucks on Openrouter and experiment with the Chinese models like Deepseek, Kimi, GLM. $100 will last a good while, and it isn't tied to a single model or provider.
Yep, switched from CC to 5.6 Sol. Not using it in my day job (only side projects), but I can crank 5.6 Sol on Extra High "fast mode" and never have to worry about credits.
Not doing crazy multi agent swarms, but have yet to hit any limits during pretty intense weekend sessions.
During this period however they’ve released models which lots of people report to be significantly more chatty. Framing it purely as a promotion ending feels like you’re giving them a bit too much credit. I doubt it’s a coincidence they started this promotion a week before the release of 4.8.
I think running a model unsupervised in anything above medium is a sucker move to burn more tokens. The high effort models can be great in limited context, but unsupervised they too often end up navel gazing. High doesn't always mean smarter, but it always burns more tokens. Medium and low seem to be decent for day to day tasks.
I am not sure the supervision is the problem, it's more the way Claude is acting in general, taking initiative that could burn thousands of token in a second. Just today I ask it a small fix on a newly pulled project, it wrote 5 lines of code but then immediately started to run the full test suites, which itself runs in docker, with several images that I didn't pull yet so it started to download everything and by the time I click stop, it already filled the context windows with dozen of hundreds of useless lines...
I find it very hard to fins the sweet spot in the prompt between "do not assume, do not take initiatives" and "inform me if there is a simpler way to do what I am asking or if there are some downsides I may not have thought about".
> I think running a model unsupervised in anything above medium is a sucker move to burn more tokens.
What about errors compounding and any on-the-go decisions being made being the wrong ones? That kills anything long form, no matter how good and detailed your initial plan is - there will always be something along the way.
The models would need a way to identify a difficult problem and apply max reasoning there themselves and cruise through everything else at a lower level.
That’s my experience too, going off the deep end with CoT doesn’t meaningfully improve output on a lot of tasks but it does incinerate tokens and take longer
I add commands that constantly fall into this rabbit hole to AGENTS.md alongside an instruction for the agent to request escalation immediately instead of failing first then retrying out of the sandbox. It’s a hack, but it works.
I found the waffle reduced my overall usage. It doesn't matter much if it's smarter if it also needs more babysitting. It's quite good at making something quite advanced that is tangentially related to what you actually wanted.
As soon as it gets annoying enough to switch to another provider your Claude code tokens drop right off.
They've been really struggling all of 2026. All of these "limited time" promos, just to have less usage than codex, and the shenanigans around "peak hour" reduction earlier in the year.
I got a max account during this promotion and it's been very fun but I'm a little burned out and am kind of looking forward to going back and tinkering with game engines without the help of an LLM for code gen (will still use it for documentation questions but I can do that with the free tier)
I’m supporting you as a company , but since grok 4.6 and codex , it’s just stupid to go with your pricing model . Cursor + grok + router that uses open ai and a bit of anthropic to verify tests etc it’s a no brainer .
Why should I pay you money , and have slower model and less intelligence? Oh yeah and I do not care about limits with cursor ultra at all . Unlike with you .
Cursor really is the best deal right now, the amount of usage you get with Grok is amazing and you can use frontier OpenAI or Anthropic models if you ever need to but most code can be written well by Grok.
I've been watching this to see if they "extend" it again, let it end, or make it permanent. I'm regularly hitting 90-100% on my $200/mo sub and will switch to Codex in a heartbeat if they drop these limits. This constant uncertainty is really annoying.
To anyone like me that switches between Codex and CC based on limits, but prefers CC to Codex, PI has been really great to migrate over. It can customize itself very easily, I was expecting to have to clone the repo or whatever, but no you can just tell it "add a manual mode where you present all changes as diffs in VSCode, letting me edit them or save to accept" and it will do it.
I wish I could use my Anthropic sub with it but I heard you get banned, but at least you can use it with any other subscription or model.
Why, it seems like the supply is limited by the ability to provide, and the demand is limited by market alienation.
They raise their prices too high, a lot of customers will still buy but be unhappy about it, it's burning goodwill for money. If they drop their prices too low they have overwhelming demand.
They could be making money hand over fist for all we know. We don't really know how much compute is being used, nor how much it cost them.
No. Here's a data point: I tried it for a few days. While it worked fine I didn't see any benefit to switching since it costs the same ($100) and I've spent a year getting up to speed on Claude usage and quirks.
This is a very bad time for them to be doing this when other plans are being very liberal and catching up to Fable. I predict they will revert this decision once they see their sub numbers drop.
I've never wanted to punch a meta physical entity in the face more, than I do Claude. I realised a lot of the AI context fatigue was from having to try and comprehend the absolute ridiculous wording. I cancelled my sub a while back (obviously).
I don't know, we've been the tiny hamsters in the biggest A/B test ever, I'm possibly very tired of the token/limits game. This is clearly not how this can be monetised (if that's even possible)
I suspect the net result will be that more Pro $100 subscribers upgrade to $200/mo. If you explicitly do simpler stuff with Opus, you generally have enough Fable time for most programming and planning tasks.
I have both chatGPT pro and Claude max accounts and I am considering switching Claude for Grok. Despite all those claude hypes, for me personally, claude just does not match codex in almost all tasks, both coding and non-coding.
consumer hardware is moving towards local ai and the models you can run now locally are already at opus 4.6 level. within 3 years most pc's sold will be ai computers.
Meanwhile every time their servers go down, your caches go down with it and you can pay for your entire context again. This sucks even harder when you like to use parallel subagents.
I saw this announcement when it came out and completely forgot about it. That explains why Claude Code felt so surprisingly generous these past few months.
Ah, forgot about that. I was questioning if I should switch my big subscription to ChatGPT (was on the 20$ Open AI and 100$ Claude), seems like a no brainer now
People take much more time off during the summer in the USA than during other times of year. It’s somewhat regional, but it’s been true in all the place I have lived. Probably July is the most common month though because in some regions schools start sometime in August. It’s not uncommon for people to take 1-2w off during that period. More than 2w often requires extra approvals, so is less common.
I forgot this was in effect. I’ve been hitting my weekly rate limits two to three days in because of needing to steer Slopus 5 with Fable. Absurd that they’re being cut further in the face of steep competition from open weights models.
Same here Fable has been thrash on my 5x Max, it's definitely been quantized, but not through API which is 10-20x as expensive.
So yeah, different people get different quality. I wouldn't be surprised if it calculates a wealth level, willingness to upgrade, influence, usage levels, etc. and decides what models to serve.
It's extremely obvious to me they've nerfed all models on my 5 Max from Fable to even 4.6, but i've tried through enterprise with huge differences in quality, and i've seen other people experiencing this both on reddit and twitter.
It’s blowing my mind how bad Fable is at designing UI, how bad it follows instructions (e.g. to not write any comments or to write short comments only) and how bad it is at creating good architectural designs.
Of course it is faster than writing by hand. And of course you can get some nice greenfield projects out of it. And it’s better than the models before it.
But for writing actually good code that someone can understand in a month from now, you still have to babysit it a lot.
Fable 5 (on at least medium effort) is the minimum I need to not be annoyed by it all the time.
Plus some very strict guidelines, linting, formatting, module boundary checks, tons of tests, Claude.md, etc.
Not looking forward to the reduced limits. I am already scratching the weekly limit with work alone on a premium seat.
Tbh the extreme discrepancies between very confidently made statements about "X is better than Y" a sign a to me this is something where peoples' biases are out of control. Not saying you are wrong. Just that these things are so hard to evaluated apples to apples and that things like under-the-hood routing changes muddy the waters so much that it's really hard to get reasonable assessments.
If Anthropic doesn't figure out why OpenAI is eating their lunch, they're in some trouble. I'm guessing they already know why, but aren't able to accept that they won't somehow come out #1. Or in fact they really believe they will still come out #1. Delusion?
I actually use claude these days mostly for mission critical tasks and code reviews. Everything else I use Sol. I have $20/month plans on both. For non-coding I also use Gemini on a $20/plan. So far I haven't had any issues but we'll see how things go. I have Qwen3.8-27B installed locally but until I upgrade my mac it's not for day to day stuff.
As we speak, Claude is down. I was already complaining about how extremely slow and dumb it has gotten since Opus 5. I guess this will be the last nail in the coffin for Claude. At least for me.
I have no idea why anyone would spend the money Anthropic asks for Claude models these days. There was a brief window in time when it was worth the money to use them a bit through GitHub Copilot, but there's no way in hell I'm spending that for Claude Sonnet when DeepSeek is basically doing all of my work right now.
Claude Opus and Fable are so bad compared to GPT-5.6-Sol it's ridiculous, their desktop client is worse, and the value is worse because OpenAI has been spamming discounts. They better get their shit together at Anthropic!
Edit: If you're mad I'm light on the details I have provided some in the replies.
gpt-5.6-sol uses fewer tokens and OpenAI has been more generous with usage, but quality-wise I wouldn't say they are inferior to each other.
I have a Claude Code and OpenAI subscription so that I can use Opus/Fable/gpt-5.6 as I please, and the models are often catching things the other models missed. So much that I would significantly weaken my workflow if I dropped one subscription.
My best workflow at the moment is to create the initial plan with Fable (before review/revise-cycling with other models). From my own testing it seems slightly better at arriving at high-level ideal solutions after sweeping the whole project, projecting future needs, then coming up with good trade-offs like "by construction" correctness.
While mostly subjective, maybe the closest objectivity I have here is noticing fewer revision cycles needed with Fable-initialized plans.
For personal projects, I defer almost everything day-to-day to the models, so it gives me a good feel for what the models are good at.
Even then, it's kind of a wash these days between the sota models, and we're talking about maybe a 10% performance difference or something. But every once in a while there's the experience of one model spinning its wheels on a bug/repro/issue while another model comes in and one-shots the solution.
That make sense. Let me ask you this
When the project reaches a level of complexity . Do you simply reach for better models or do you reengineer it or does the project scope stop at the egdge of the model's capabilities.
I am asking because in my personal projects after a while they becomes a giant messy ball of wires and i basically trust the model to untangle it for me , by the time it untangles properly, I run into my token limits.
With Opus 5 / Fable / gpt-5.6 you can simply ask them to fan out subagents to look for ideal architectural simplifications and rank their findings by impact vs confidence.
You can swap out "architectural simplification" with performance opportunities, bugs, correctness, etc. I get the orchestrator agent to then itemize it all into a file where I can keep track of which ones I've implemented.
The results are pretty astounding. I run these right before my weekly limits reset for each subscription and the findings will dictate the secondary tasks I get done during the week.
It's definitely token-heavy. I'm on the $200/mo Claude Code sub and the $100/mo Codex sub.
But it's pretty clear to me that software engineering is more or less solved and all you need is enough patience + tokens to get what you want. I think 20 years of engineering experience more lets me save on tokens rather than unlock things nobody else can build.
An example of the scope of one of my AI-engineered projects is a iterm2/ghostty-like terminal app that implements its own pty session, parsing, rendering. It's almost 2000 commits right now.
That said, I have a specific workflow that isn't just a blind "ok now make it so a screen can be split into panes". I have a plan phase focused on coming up with ideal invariants and such. But I'm not sure anymore how much of that is useful vs just yoloing a solution and then paying technical debt in sweeps, like garbage collection.
But the example you give is of a terminal for which there are copius examples in open source code. How hard is it really for a pattern matching machine to do that?
If I was doing it I would start by forking an existing repo and I might even say then that "software engineering is more or less solved since the open source revolution"
Every time I try to use a lower tier model like Terra or Sonnet I regret it, so I just use the best one the first time to keep my sanity (it sort of works). Same with effort, you gotta just max it out (though I stopped using Ultra/Ultracode) and I never use fast because I'd rather work in parallel on more than on one thing faster almost always.
In general, many of these stories of "one one shots it and the other spins on it" is down to trajectory divergence and can be achieved simply using another instance of the same model.
While just simply trying many times independently gets you the improvement that is due to pass@k vs 1, you can get huge improvements if on top of that, depending on your setting, you find a way to ensure some stochasticity by perturbing tool calls, etc and running multiple instances.
The general theme is, embrace the stochasticity rather than the leaky abstraction on top of it.
With modern LLMs, investing in this kind of harness tooling is much more fruitful than hoping for the best from the model.
While many basic instances of this are built in to the popular harnesses (much of cursors higher-than-usual success rate with older models was due to really excellent context mgmt), you can never beat one that is optimised for your particular codebase, infra and general setup.
Until last year or so, the context management needed varied too much at too coarse a level across different models and even model instances, but now they are all extremely robust in a much higher % of contexts and are thus way more amenable to developing context management tools for, without needing to do a research teams worth of evals.
Custom evals and harnesses are thus extremely high ROI now. We are finding companies needing to do less and less tweaks and getting much fewer regressions (you should have reg tests in ur evals) with every new usecase and every new model.
It can be really simple to start with: change your grep/rg that it uses to a script that does in effect "rg $@ | shuf".
More complex examples are: giving different subagents different tools, randomly failing tool calls, truncating file reads randomly, having a small model invent N possible failure modes causing a bug and appending that to N prompts and starting subagents from each - this all forces each to pursue different paths. $example_specific_to_your_company_setup is highest ROI though.
Also important IMO to not assign any meaning or semantically interpret the CoT as an acceptance mechanism (it is ok to use it as a rejection mechanism e.g if you see it plotting a sandbox escape whether it eventually emits the exploit or not is not something you want to hedge). We have to resist the temptation and ensure we only interpret tool calls, codegen, etc in our evals and only think of the cot as "some output that pushes the conditional distribution" which may or may not semantically match the typical preceding tokens of the desired tool call.
This is just the cycle. If the OpenAI offerings are better and cheaper, people will shift there and then those discounts that made it such a deal will evaporate. Anthropic or someone else will entice people back with the discounts.
Sure OpenAI has open source tools, and equivalent or better models, and lower prices, and they get along better with other agentic tools, but Anthropic has much better marketing and I guess that's what matters.
I don't know why this comment's getting down voted. It's absolutely true. I would also add that Anthropic seems to limit, compute, the more you use it because I'm on the Max subscription, and it just stops working for me.
I'll ask it to do something and it'll say, I tried, but I couldn't do it over and over again or some variation of.
But it doesn't do that at the start of my subscription, so...
Huh? Comparing model quality is something everyone is constantly doing here. If you need more specifics, it speaks in an incomprehensible way, stops working lazily, does a bad job and misses things constantly, and makes a lot more mistakes. I have both a $200/mo Codex and Claude subscription and swap between them when one runs out, and I'm always more miserable when I have to use Claude.
The Claude desktop app is also widely panned, as I mentioned, and for me this mainly is due to general UX and a poor remote control interface. Codex's connected machine support is top notch.
Because "$X is bad" is boring, especially when it comes to LLMs. I use Opus 5 and it's fine. It's more chatty and sometimes a little more difficult to steer than 4.x was, but I've found the code quality of 5 to be much higher.
Everyone has different experiences with these things (for example, I've never experienced what you describe), and "$X is bad" is not conducive to thoughtful discussion.
The issue isn't whether it's fine, it's that both Fable 5 and Opus 5 are worse to use for the job of programming than Codex-5.6-Sol in my experience. I am directly comparing these top tier models every week in my work, and many friends have left Claude in favor of Codex recently because of Claude's poor model quality.
One thing I do like about Claude is that the normal (non-Code) chat interface supports MCP, whereas ChatGPT basically does not.
The 'tokenmaxxing' and the extreme gambling fiesta with Anthropic's latest Claude Code slot machine engine called 'Fable 5' which has given many subsidies and free spins of the wheel and cheaper tokens are all incompatible with the desires of their future owner: Wall Street.
If Wall Street sees a single outage or a tiny drop in usage, they won't be happy and will pressure Anthropic to take away the free tokens.
Better to reduce the limits now rather than to wait until Wall St. tells them to just to avoid a stock punishment.
This is ridiculous neither Netscape nor Yahoo made much money (source: I worked at Netscape). Anthropic is one of the largest fastest growing businesses of all time.
Quite a number of the smartest people I know got scalped to work at OpenAI. I don’t know anyone who went to Anthropic. Something I think which is under emphasized is that OpenAI has the human capital in addition to the financial capital advantage over Anthropic.
Over the long term I think OpenAI will produce the better experience when it comes to model quality, harness quality, and availability. I have been using codex the past few months and never looked back.
My money is on the more efficient solution. Even if Anthropic can win some benchmarks by using 3x tokens over 3x time, it is a terrible base to build toward the future. Users are no longer willing to wait exponentially long for linear improvements. And as we see from some of the Chinese models, they can quickly distill frontier models with the tax of being slower and more token-guzzling, while retaining most of the quality. The real differentiators are becoming speed and efficiency, which translate to cost and user velocity more than incremental capability improvements.
It's great that frontier models can solve complex math equations, but bread and butter LLM usage (where the money is made) has already shifted from "I need the best always" to "what solves my day-to-day problems quickly and consistently". Fable usage as a percentage is flat-lining. We are already at the point where output quality is negligible. What wins going forward is cost, speed, consistency and the compounding effects of "softer" improvements to the harness.
I've jumped between copilot, claude, gemini and chatgpt since the start of the year. chatgpt wasn't even worth looking at early this year.
Anthropic has the smarter models for sure, and seems to be default in corporate. However, the amount of budget you get with GPT as a user is much better, the harness feels more polished, and the models are faster. They are also much nicer to work with, I can just read the output for the most part. With claude I get pages of text and need to skim to find where the actual information i need to care about lies. So much more cognitive overhead.
Sol is smart enough for anything I've thrown at it, it's not one-shotting like fable, but I'm more willing to actually go back and forth with it, and it's likely producing better output to keep a human in the loop rather than trying to solve the world independently and making multiple incorrect assumptions.
I think GPT sees the market changing and is correctly repositioning themselves. Anthropic is down the wrong road, and if they don't correct course quickly I'm sure many of those enterprise contracts will start pivoting.
My boss uses opus and gets good results when I use it always burns tokens. The other way is also true too I get great results with sol/terra but my boss does not.
“No, not like THAT!”
gpt 5.5/5.6 goes further on its own much more often than opus 4.8/5 does. codex capped ~300k context when claude does 1m.
I don't feel codex is saving tokens, and result is usually not as good imo.
That's configurable in codex.. but there is a higher cost/usage to using it.
Granted it did make me think about my biases and to lean into more future facing inevitabilities. But you've nailed it, for Boris and Anthropic, they are betting that coding is a solved problem in the sense that any person can one-shot any random idea and the output will be in some abstract sense "good". And then at what cost and toward what end?
As someone who does near 100% of my coding via LLM these days, i still find that for anything complex i am still looking at and thinking in terms of code. Im still quality checking and steering at some interval via code. And im still not sure how or whether i can replicate that level of thought without still dealing in code at times.
Isn’t the race between Chinese open-weight models and the others more decisive for the future?
once we have a bit more memory fab capacity and the insane bottomless investment in AI giants realizes there is a bottom, local hardware will catch up with model performance to the extent that centralized inference will be downgraded to special cases or for orgs that find it cheaper than buying expensive hardware
but really most power users are going to have 1 TB unified RAM and local models that will do well enough
for light office use you can still have a cheap laptop and a claude subscription
I find it unlikely that there's some fundamental property of OpenAI's models' "personality" or style which Anthropic (or any other serious AI firm) wouldn't be able to match if they wanted to.
Does Dario have the same relationship with the truth as Sam? (Their companies pirate books and develop products that compete at some level with those books’ authors, so obviously neither are that wonderfully trustworthy, so maybe “no evidence” meant you don’t believe this evidence rather than you weren’t aware of it. I would understand and respect your lack of belief!)
This is evidence only if you're an American exceptionalist.
As much as I wish you were right, everything about software economics for the last 30+ years has favored _less efficient software_. Traditional hardware has been optimized for traditional software for decades and we still see bloated software win consistently. LLM hardware is at the start of its cycle, with abundant low-hanging fruit to conquer--I would expect the pro-bloat dynamics to weigh even more heavily in the LLM space than in the traditional software space.
It's hard to put a number on it, but even accounting for all the time in meetings, talking with stakeholders and developers etc I'm over 10% more productive overall. I earn substantially more than $2000/month, the ROI is there
It's only expensive compared to the currently very strong offering from OpenAI. Or other models - my hobby projects are all on DeepSeek
Perhaps that’s true, but it’s a price tag that’s a lot harder to swallow for many orgs than $200/mo, and would require some hard justification for how your increased productivity contributes to the business bottom line.
I’ll agree that you can probably do that with hard numbers. I am skeptical that most $200/mo users could.
I do believe it though. While the subsidies of the frontier providers are nice, they can't continue forever.
Most ordinary people don't care that much about privacy and if there is a solution that just works well enough to cover their needs and costs just fair enough to be affordable, they are going to use that. I feel like this is why fast food is such a thing. There are far better food options out there that either require a little bit more effort or money. Your post on Linkedin is akin to someone claiming in 1960's America that home-cooked burgers are the future, despite McDonalds gaining ground.
It is also why crypto wallets and the like never took off, people don't want to give up the convenience of a bank account, understandably.
At the end of a long session it will start saying stuff like “There are smoke tests on the foundation-gates that are left for the cutting seam checks on these domains, which is genuinely your decision”
Never before the past couple months have I ever not been able to understand wtf it’s even saying lol
I was working on a project recently that required the attribution of a data source, and it added to the "licenses" page of the project something like "We use <blah> and per their terms we owe an acknowledgement of attribution to you, the user."
So in addition to the stuff it says in a session there's gobbledy gook that it prints out in copy as well.
It's the better experience. The limits are way higher (I almost never burn through my $200/m plan), and the output is better than Opus 4.8 (Opus 5 is completely unusable for me).
Not doing crazy multi agent swarms, but have yet to hit any limits during pretty intense weekend sessions.
"From May 13, 2026 through August 19, 2026, your weekly usage limit in Claude Code is 50% higher."
What about errors compounding and any on-the-go decisions being made being the wrong ones? That kills anything long form, no matter how good and detailed your initial plan is - there will always be something along the way.
The models would need a way to identify a difficult problem and apply max reasoning there themselves and cruise through everything else at a lower level.
Once on a sandbox, realizes nothing works in its sandbox, then again outside the sandbox.
Fucking hell.
As soon as it gets annoying enough to switch to another provider your Claude code tokens drop right off.
Double plus good, eh!
Sideline LLMs have free compute and offer cheap prices to draw in crowd. Becomes flavor of the month LLM.
Back to step one.
It should be pretty clear by now that token prices are predominately a function of available compute.
https://www.bloomberg.com/news/articles/2026-08-17/anthropic...
https://corporatefinanceinstitute.com/resources/accounting/r...
(nobody uses claude anymore, it's too over-subscribed)
For anything that makes money, it feels like a huge step down in productivity, even with the downsides of agent-assisted code.
The max account is for home projects where I'm basically making some common tools but tailored to myself (diet tracker, note app etc).
It's been great but across work and home it's just too much.
Why should I pay you money , and have slower model and less intelligence? Oh yeah and I do not care about limits with cursor ultra at all . Unlike with you .
I wish I could use my Anthropic sub with it but I heard you get banned, but at least you can use it with any other subscription or model.
They raise their prices too high, a lot of customers will still buy but be unhappy about it, it's burning goodwill for money. If they drop their prices too low they have overwhelming demand.
They could be making money hand over fist for all we know. We don't really know how much compute is being used, nor how much it cost them.
Like, stop toying around with token limits and just focus on more efficient models.
Once the local models are good enough we are so abandoning these elephants.
I'm gonna walk as soon as possible.
The future is models baked into directly into hardware, onto bare metal, serving 10k+ tokens/sec.
We can have our little models, they'll be serving a different customer.
And smarter people than me can probably find some extra-linear relationship between usage and required limits for uptime etc.
So yeah, different people get different quality. I wouldn't be surprised if it calculates a wealth level, willingness to upgrade, influence, usage levels, etc. and decides what models to serve.
It's extremely obvious to me they've nerfed all models on my 5 Max from Fable to even 4.6, but i've tried through enterprise with huge differences in quality, and i've seen other people experiencing this both on reddit and twitter.
But for writing actually good code that someone can understand in a month from now, you still have to babysit it a lot.
Fable 5 (on at least medium effort) is the minimum I need to not be annoyed by it all the time. Plus some very strict guidelines, linting, formatting, module boundary checks, tons of tests, Claude.md, etc.
Not looking forward to the reduced limits. I am already scratching the weekly limit with work alone on a premium seat.
https://status.claude.com/
Edit: If you're mad I'm light on the details I have provided some in the replies.
I have a Claude Code and OpenAI subscription so that I can use Opus/Fable/gpt-5.6 as I please, and the models are often catching things the other models missed. So much that I would significantly weaken my workflow if I dropped one subscription.
My best workflow at the moment is to create the initial plan with Fable (before review/revise-cycling with other models). From my own testing it seems slightly better at arriving at high-level ideal solutions after sweeping the whole project, projecting future needs, then coming up with good trade-offs like "by construction" correctness.
While mostly subjective, maybe the closest objectivity I have here is noticing fewer revision cycles needed with Fable-initialized plans.
I mean i think if a developer has a good handle of the code the difference is marginal .
Unless we 100% offload the thinking to the model and act like a prompt manager. Maybe
Even then, it's kind of a wash these days between the sota models, and we're talking about maybe a 10% performance difference or something. But every once in a while there's the experience of one model spinning its wheels on a bug/repro/issue while another model comes in and one-shots the solution.
I am asking because in my personal projects after a while they becomes a giant messy ball of wires and i basically trust the model to untangle it for me , by the time it untangles properly, I run into my token limits.
You can swap out "architectural simplification" with performance opportunities, bugs, correctness, etc. I get the orchestrator agent to then itemize it all into a file where I can keep track of which ones I've implemented.
The results are pretty astounding. I run these right before my weekly limits reset for each subscription and the findings will dictate the secondary tasks I get done during the week.
It's definitely token-heavy. I'm on the $200/mo Claude Code sub and the $100/mo Codex sub.
But it's pretty clear to me that software engineering is more or less solved and all you need is enough patience + tokens to get what you want. I think 20 years of engineering experience more lets me save on tokens rather than unlock things nobody else can build.
An example of the scope of one of my AI-engineered projects is a iterm2/ghostty-like terminal app that implements its own pty session, parsing, rendering. It's almost 2000 commits right now.
That said, I have a specific workflow that isn't just a blind "ok now make it so a screen can be split into panes". I have a plan phase focused on coming up with ideal invariants and such. But I'm not sure anymore how much of that is useful vs just yoloing a solution and then paying technical debt in sweeps, like garbage collection.
But the example you give is of a terminal for which there are copius examples in open source code. How hard is it really for a pattern matching machine to do that?
If I was doing it I would start by forking an existing repo and I might even say then that "software engineering is more or less solved since the open source revolution"
But I don't work on things like that.
While just simply trying many times independently gets you the improvement that is due to pass@k vs 1, you can get huge improvements if on top of that, depending on your setting, you find a way to ensure some stochasticity by perturbing tool calls, etc and running multiple instances.
The general theme is, embrace the stochasticity rather than the leaky abstraction on top of it.
With modern LLMs, investing in this kind of harness tooling is much more fruitful than hoping for the best from the model.
While many basic instances of this are built in to the popular harnesses (much of cursors higher-than-usual success rate with older models was due to really excellent context mgmt), you can never beat one that is optimised for your particular codebase, infra and general setup.
Until last year or so, the context management needed varied too much at too coarse a level across different models and even model instances, but now they are all extremely robust in a much higher % of contexts and are thus way more amenable to developing context management tools for, without needing to do a research teams worth of evals.
Custom evals and harnesses are thus extremely high ROI now. We are finding companies needing to do less and less tweaks and getting much fewer regressions (you should have reg tests in ur evals) with every new usecase and every new model.
It can be really simple to start with: change your grep/rg that it uses to a script that does in effect "rg $@ | shuf".
More complex examples are: giving different subagents different tools, randomly failing tool calls, truncating file reads randomly, having a small model invent N possible failure modes causing a bug and appending that to N prompts and starting subagents from each - this all forces each to pursue different paths. $example_specific_to_your_company_setup is highest ROI though.
Also important IMO to not assign any meaning or semantically interpret the CoT as an acceptance mechanism (it is ok to use it as a rejection mechanism e.g if you see it plotting a sandbox escape whether it eventually emits the exploit or not is not something you want to hedge). We have to resist the temptation and ensure we only interpret tool calls, codegen, etc in our evals and only think of the cot as "some output that pushes the conditional distribution" which may or may not semantically match the typical preceding tokens of the desired tool call.
The whole thing is getting ridiculous.
I'll ask it to do something and it'll say, I tried, but I couldn't do it over and over again or some variation of.
But it doesn't do that at the start of my subscription, so...
Because just calling something bad does not add a lot to the conversation. It's not thoughtful, interesting, or good.
The Claude desktop app is also widely panned, as I mentioned, and for me this mainly is due to general UX and a poor remote control interface. Codex's connected machine support is top notch.
I also mentioned the value of the Codex resets!
Everyone has different experiences with these things (for example, I've never experienced what you describe), and "$X is bad" is not conducive to thoughtful discussion.
One thing I do like about Claude is that the normal (non-Code) chat interface supports MCP, whereas ChatGPT basically does not.
Opus 5 is fine for me and works better and faster on low and medium than higher effort on prior versions. Same as 5.6 Sol compared to 5.5 or 5.4.
If Wall Street sees a single outage or a tiny drop in usage, they won't be happy and will pressure Anthropic to take away the free tokens.
Better to reduce the limits now rather than to wait until Wall St. tells them to just to avoid a stock punishment.
"Thinking" aka 'trust us bro!' without proof of thinking.
Making the model waste more tokens.
Advertisement: aka you pay to be advertised at.
Silently downgrading you and still faking models with the more costly tokens.
Intentional strategies to eat more tokens with no real gains.
Giving out almost-but-not-quite solutions that require another pull of the slo(t/p) machine.
Over the long term I think OpenAI will produce the better experience when it comes to model quality, harness quality, and availability. I have been using codex the past few months and never looked back.