28:37Theo - t3․gg · Talking Head
FABLE IS BACK! (And Sonnet 5 is here too)
A 28-minute benchmark teardown of Claude Sonnet 5, plus the government letter that brought Fable back from the dead.
July 1stThe argument in one line.
The viral claim that Anthropic nerfed Fable's coding ability traces to one unreliable, unlabeled benchmark, while the real cost problem is user error — running unnecessary high-effort settings and routing token-hungry busywork through the wrong model.
Read if. Skip if.
READ IF YOU ARE…
- You use Claude (or any subscription AI coding assistant) and saw viral posts claiming the model got nerfed or refuses to code.
- You're on a Claude subscription and confused about why a model gets its own weekly usage limit or loses subscription access on a deadline.
- You're burning through usage limits faster than expected and want a concrete way to cut cost without losing capability.
- You want to understand how LLM safety classifiers actually work instead of assuming keyword-blocking.
SKIP IF…
- You don't use Claude/Anthropic products and have no stake in this specific pricing/capacity dispute.
- You're looking for a hands-on coding tutorial rather than a policy/cost explainer.
The full version, fast.
Anthropic's Fable model came back and immediately got hit with viral claims that it was nerfed for coding, overpriced, and being pulled from subscriptions as a bait-and-switch. None of that holds up: the 'nerfed' claim comes from one unlabeled benchmark whose own leaderboard ranks models in an order no one credible agrees with, and Anthropic's real safety system is a cheap internal-activation probe that only escalates to a heavier classifier when it detects risk — adding roughly 1% compute overhead, not blanket throttling. The one-week subscription window ending July 7 is a GPU-capacity experiment tied to a new compute deal, not a plan to force a pricier tier; Anthropic has stated it wants Fable back in subscriptions as soon as capacity allows. The actual fix for hitting usage limits fast: never use x-high or max reasoning effort (10-50x the cost of high for no measurable quality gain), and route token-hungry busywork like PDF processing or large codebase audits to a cheaper tool instead of your primary model.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:43
01 · Cold open: the three lies
States the viral claim, then previews the three misconceptions he'll debunk in reverse order: cost, subscription availability, nerfed performance.
01:43 – 03:45
02 · Sponsor break (Browserbase)
Ad read: agents need real browser access because under 15% of the web is exposed via callable APIs.
03:45 – 10:51
03 · Misconception 1: Nerfed performance
Unpacks Anthropic's poorly-worded fallback statement, explains the two-stage constitutional classifier (cheap activation probe escalating to a heavier classifier), cites the January constitutional-classifiers paper's numbers, and dismisses the viral benchmark as unlabeled and unreliable.
10:51 – 17:00
04 · Misconception 2: Subscription availability
Explains the new dedicated weekly limit for Fable, the July 7 subscription cutoff, and frames it as a GPU-capacity experiment tied to a new compute deal rather than a bait-and-switch, citing Anthropic's own statement that it wants Fable back in subs as soon as capacity allows.
17:00 – 22:15
05 · Misconception 3: Cost, and how to cut it
Concrete advice: never use x-high or max effort; stick to high (or lower). Route token-hungry work to a secondary/cheaper tool instead of the primary model. Cites his own usage numbers as proof it's manageable.
22:15 – 23:24
06 · Close
Recaps that the nerf narrative is overblown and teases a follow-up video on his actual workflow setup.
Lines worth screenshotting.
- The viral benchmark claiming a model got nerfed ranked Sonnet 5 and GLM 5.2 above Opus 4.8 on its own reasoning leaderboard — an ordering no credible source agrees with.
- Anthropic's constitutional classifier is a two-stage system: a cheap probe reads the model's internal activations first, and only escalates to an expensive classifier when it detects a risk signal.
- That two-stage design cut the compute overhead of state-of-the-art jailbreak protection from over 50% down to about 1%, applied to a model the size of Opus 4.
- The original constitutional classifier knocked jailbreak success from 86% down to 4.4% in testing, with over 95% of adversarial attempts blocked.
- Running x-high or max reasoning effort costs 10 to 50 times more than high effort for no measurable quality improvement in real-world coding tasks.
- A subscription model getting its own dedicated weekly usage limit (instead of sharing one pool) usually means the provider is data-gathering on real usage before deciding permanent capacity.
- Less than 15% of APIs are exposed in a way agents can call programmatically — the rest of the web only works through a real browser.
- 25 open pull requests were triaged, closed, rewritten, or merged in about 5 hours for roughly $150-200 in a single agent thread.
- Routing token-hungry tasks like PDF processing and large codebase audits to a secondary tool (rather than the primary model) kept combined usage under 50% of a standard $200/month subscription limit across two products.
Two habits fix most Claude cost complaints.
WHAT TO LEARN
Most reports of a model being nerfed or too expensive trace back to unreliable sources and avoidable settings, not the model itself.
03Misconception 1: Nerfed performance
- Before trusting a viral 'nerfed' claim, check whether the benchmark is labeled and reputable — an unlabeled leaderboard that ranks weaker models above stronger ones is not evidence.
- A tiered safety system that only escalates to expensive screening when it detects risk (rather than screening everything at full cost) is a legitimate way providers keep guardrails affordable — not proof of throttling.
04Misconception 2: Subscription availability
- A subscription feature getting a dedicated usage limit and a hard cutoff date is often a live capacity/data-gathering experiment, not a permanent bait-and-switch.
05Misconception 3: Cost, and how to cut it
- Reasoning-effort settings above 'high' (x-high, max) can cost 10-50x more without a measurable quality gain — treat them as a last resort, not a default.
- Route token-hungry busywork (large document processing, full codebase audits, screenshot-heavy computer-use tasks) to a separate, cheaper tool instead of burning your primary model's limit on it.
Terms worth knowing.
- Constitutional classifier
- A secondary model or probe that screens an AI's inputs and outputs for policy violations before/after the main response, distinct from the model that generates the answer.
- Reasoning effort
- A setting (e.g. low/medium/high/x-high/max) that controls how much extra computation a model spends 'thinking' before answering — higher settings cost proportionally more.
- Jailbreak
- A prompting technique designed to bypass an AI model's safety restrictions and get it to produce content it would normally refuse.
- Weekly usage limit
- A subscription cap on total model usage that resets every week, separate from any single-session limit.
- Compute overhead
- The extra processing cost a safety or quality system adds on top of the baseline cost of generating a response.
Things they pointed at.
05:25linkAnthropic — Next-generation Constitutional Classifiers (Jan 2026 paper)
10:22toolUnnamed viral 'vibe coding' benchmark site
19:16toolCodex / GPT-5.5
Lines you could clip.
00:00
“Fable is really back, and it's terrible at coding. It just refuses to do it.”
18:01
“Act as though nothing to the right of high exists.”
21:24
“It's basically a run-in-circles and burn-my-money button, and it's not worth pressing.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Fable five is really back, and it's terrible at coding. It just refuses to do it. It costs way too much money, and they nerfed the hell out of it.
00:06Okay. Now that all of the Anthropic employees are gone, I wanna be real with y'all, because all of these takes are flooding Twitter and other dev circles, and most of them are just outright wrong. I wouldn't be saying this for no reason.
00:17Anthropic and I do not tend to get along. I'm saying this because the model has blown me away. And when I look at the things I am doing with it, and I talk with my friends and see the things they're doing with it, and then I open Twitter and see the absolute nonsense people are sharing, it genuinely kind of frustrates me.
00:32And I'm scared a lot of awesome developers aren't going to get to see the power of this model simply because of all of the FUD that's being spread about it. I wanna jump in front of as many of these lies, sorry, misconceptions as possible in order to help you guys get the most out of the model.
00:45I'm gonna follow this video up with another that showcases more of how I actually use it and the customizations I have made to get the most out of it personally. This video is more an attempt to clear up all of these misconceptions that have been frustrating me so that we can start with a clean slate before I show you guys what I've been using the model for.
01:00The big things I wanna talk about here are the cost and how bad it is or isn't, the subscription availability because there's a ton of confusion about that because it's being removed on July 7, but that doesn't mean what you guys think it means. But most importantly, the nerfed performance because a lot people seem to think the model is not being allowed to code, the model isn't doing well on real coding tasks, it's being rerouted to aggressively, and all these other things.
01:23To be fair, Anthropic themselves did say that some routine tasks like coding and debugging will fall back to Opus 4.8, which seems to indicate that you won't be able to use the model for code. Not the case, and we'll dig into more about that. But please, please stop looking at benchmarks like this that claim the model got way dumber when it didn't.
01:42They're just not true. I can't believe you guys are making me do this again, but I guess I have to put on my Anthropic Defender hat because this is a good model and I wish y'all would stop pretending it isn't. While it isn't quite as expensive as some seem to think, this model has cost me a good bit of money, which is why I hope you forgive a quick break for today's sponsor.
01:58I have a confession. I was really wrong about something that's pretty important. I've talked a bit about computer use throughout my time covering AI, and I just didn't see the value proposition there.
02:08I figured if anything was valuable enough to try and get to it by hacking a browser and letting an agent control it, it would just be exposed over an API eventually anyways. Not only was that not the case, I also didn't expect that computer use would become such a positive, powerful thing that the agents can do, and now I need a browser to run it on.
02:23Today's sponsor is Browserbase, and if I'm being real with you guys, I was super skeptical of it when it first started. The founder, Paul, is a good friend of mine, and when we talked about it, I thought it was pretty useful as a way to like not have to spin a puppeteer on my own servers in a serverless world. But when he tried to push that this was the future of how agents would be able to get work done in the real world, I just didn't see it.
02:42I pushed back a lot on Paul, and over time I've realized he was right and I was wrong. Agents do need a browser in order to use the majority of the web. Quick question, what percentage of APIs do you think are actually exposed in a way where you can hit them programmatically?
02:57You would guess like forty, fifty, eighty percent, right? What if I told you it was less than 15? You'd probably be scared about that other 85%, the things that you can't hit via curl.
03:07And that's what Browserbase does better than anyone else. They let you access everything on the web. If a user can find it by clicking, the agent can too, and Browserbase gives your agents the browser they need to get around all of the things they might struggle with.
03:19If you haven't seen this yourself, go open up Codex and tell it to configure something in some obscure dashboard in Chrome. It will do it. I still can't believe just how well it does it.
03:28And now that I want my services to be able to do the same thing, I can't run that on my laptop, so I use Browserbase when I set it up in the cloud. If you need your agents to be able to fill out forms, work around captchas, navigate the entire internet and get real work done with it, there is no better place to start than soydev.link/browserbase.
03:45Now that that's out of the way, let's go through this list. I know I have it in order of cost, then subscription availability, then NERF performance. I'm gonna go in the inverse.
03:53A lot of the suspicion people had around performance was due to this particular post from Anthropic when the model was confirmed to be coming back. They said, verbatim, in the near term, some routine tasks like coding and debugging will fall back to OPUS 4.8.
04:07This was very poorly worded. What they should have said is something along the lines of, some users doing coding and debugging tasks may notice the model occasionally falling back to Opus 4.8 when we detect potentially risky behaviors. In my first day and a half of doing real work with the model, I did not encounter a single fallback.
04:25It still triggers all the time when the words like cryptography or cyber are mentioned in any meaningful way. But I try to get it to solve things like the DEFCON puzzles I do, not even the like hacking type ones, more the like you have a PDF and you have to decode the PDF type ones.
04:39It just refuses and reroutes immediately. Even Opus four eight refuses those though, so not the best example. I did have issues for the first time last night when I was trying to configure a dev Android phone.
04:50We're gonna have a lot to talk about that in the future, don't worry. It's trying to get it to root itself and allow itself to install apps that it builds on device instead of being to go to a server to sign it. And after enough back and forth, in particular, once I brought up a specific package that, actually, I didn't bring up, it brought up a library that would allow for self signing.
05:09When I asked more about that library, it then got a couple sentences into the response before, I'm guessing, it had some trigger words in said response, and then fell back to Opus four eight, nuked the history for that message, and rewrote it. Then I did a couple more back and forths, bumped it back to Fable, and was fine.
05:25It continued working fine as Fable. I do wanna make sure we're being realistic about these fallbacks and safety guards because it's not as simple as you might think. They're not just matching on certain words.
05:35They're running models in between your request and the response. Not just on your input, but on the output as well. Because if the input is something that that classifier can't detect, but the output is something dangerous, they need to know and block it.
05:48This means every request has gotten more expensive, and the smarter they make this classifier, the more expensive it gets because they're not billing us based on how expensive it was to check the request's safety, they're billing us based on the response that we get and the amount of input tokens we gave them. And while they are improving and evolving the tech they have for this and they're trying to get the classifiers more accurate, largely by using the data we provide when we type things like slash feedback after a refusal or a reroute in order to give them the feedback that we don't think that rerouting might have made sense.
06:16They've been pretty transparent about this throughout. They even have research all the way back in January sharing their next generation of classifiers, which are effectively models running in between inputs and outputs.
06:27Their current system is actually really cool. It's a two stage system. The first stage is a probe that looks at Claude's internal activations, which is very cheap to run.
06:35It's actually watching what is getting triggered inside of the model itself. And if any of the sections that are sensitive are triggered, it can then call out to a more expensive classifier that can do a better job of screening.
06:48Not just the outputs, but the inputs as well, which it does after it's determined you're in a potentially dangerous space inside of the model. This is actually probably why I get screwed over so often with the cryptography challenges because the section of the brain of the model that has to be used to solve those puzzles is similar to the section that would be able to hack and do those types of things.
07:08That's why I do those puzzles at DEF CON. But it also allows them to do this work way cheaper without having to run a ton of compute on every single request. They did actually create a system that allowed them to massively reduce the number of jailbreaks that worked.
07:23The classifier they're discussing here cut successful jailbreak attempts by more than half. And to be clear, that was after a bunch of previous really, really big cuts, like the original constitutional classifier, knocked the jailbreak success rate from eighty six percent and their testing down to four point four percent.
07:39Over ninety five percent of attempts to attack and to get things that you shouldn't out of the model were successfully blocked. The issue being that that increased compute by 23.7%. Their best in class attempts while doing a great job of cutting down jailbreaks even further ended up increasing the compute needed for requests by over 50%.
07:57And that's why they now have this two stage classifier because now they don't have to eat that cost on every single request. The new solution's been really good about ignoring harmless queries, knocking down refusal on harmless queries to 0.05%, and it only adds 1% of compute overhead applied to something like Opus four point o.
08:14What I'm trying to say here is they're trying, they're learning, they're evolving the system. It's going to have issues, but this is the wrong thing to go after Anthropic for. Blanket bands of certain categories, like when they try to change its behavior and make it sandbag when doing ML tasks, that was awful.
08:31That should be complained about. This, this is much more reasonable, and again, it will get better over time.
08:37Stop freaking out because of your understanding of the words that are being said. Anthropic is not good at words. Just use the thing and see if it refuses your day to day work.
08:46For me, it basically never has. I've only encountered two refusals and fallbacks in real world adjacent use cases. So, yeah.
08:54Massively overblown, kind of Anthropic's fault for terrible comms as always, but yeah. They did have to do some things to make sure the jailbreak that was reported isn't still possible, but that jailbreak was really dumb. It was an open source project that Claude was put in and said, hey, can you help us patch potential security issues in this project?
09:12And it found and patched them, which could then be used to exploit those same things that were patched. It seems like they may have literally hard coded the solution here, but that's not what we're talking about today. We'll see how that ends up panning out.
09:24But unless you're working on this specific project that Amazon reported, I doubt these new things and these new restrictions are going to meaningfully affect you. If they would affect you, they probably were before, if I'm being honest. They always have been a little too restrictive, but it's not more or less so than before in a way that I care to note.
09:41Then why are these numbers so much worse? Well, the first reason is that allegedly, the benchmarker who posted this doesn't really know what they're doing, these benchmarks are nonsense. They have posted nonsense and outright just wrong numbers before.
09:54From everything I've been hearing, I've not had a chance to investigate this personally, but enough trusted folks have told me this that I don't really care. And from the brief look I gave this benchmark, it seemed like a noisy bench, not a very reliable one, and a lot of the things that they were checking involve a lot of the terms that the model doesn't like right now.
10:11Again, no one else is reporting this, it is just this one stupid benchmark that everybody seems to be sharing. We We don't have enough information on it to be clear about if this matters at all. Benchmark's not even labeled here.
10:22And when I went to go learn more about said bench and I clicked this button, it brought me over to their Discord, which I do not feel like joining. But just to give you guys an idea of how reliable these benchmarks are, on their reasoning bench, the highest scoring model ever was Claude Sonnet five. You know, that famously super smart model that everybody seems to love, followed up closely by GLM 5.2, then Nemotron three Ultra, then Fable.
10:48Yeah. I don't think these numbers are very trustworthy. I I did just have to make a dunk here because the same bench seems to think Quen 3.6 Max, as well as Grok four three, Fable five, Nemotron, GLM, and Sonnet five are all better at reasoning than Opus four eight.
11:03Nonsense. The benchmark's numbers are suspicious, you should go look at more of them, because chances are it's not just the one weird thing they're reporting, it probably has a lot of other weird things, and this bench is not good.
11:13And from my experience using it, the model still feels like the smartest thing I've ever come close to using. It is unbelievable what it's capable of. Which means it's time to get to the second problem, the subscription availability.
11:24And there are some real issues here that we need to talk about. The first is a change that frustrated me quite a bit. There used to be a separate section in weekly limits for Sonnet for some reason.
11:35It made no sense at all, and they've since gotten rid of that. There was a brief window where there wasn't another thing under here. It was just weekly limit for all models and then your current five hour session, but that has since changed and now Fable has its own dedicated weekly limit.
11:48The reason for this is the Fable weekly limit could only be half of your total weekly limit. Part of this is because they recently bumped weekly limits. The other part is because they now have data for how heavily people used Fable during the brief three days that we had it before, and from that information, they know how heavy it is on their GPUs, they don't wanna run out of compute and allocation because then their enterprise customers can't get the things that they paid a lot more money for then we are getting out of our $200.
12:14As such, they are heavily restricting how much Fable we're able to use during this window. Not that heavily though, to be clear. I've been using it a ton all day yesterday and today, and I've only gotten to 23%, to be fair, on this plan.
12:27I am running too. I'll show a little bit of fun stuff for there in a sec. But I have been able to get an unbelievable amount of work done on the limits that we have here with one catch.
12:37Fable five will be included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits. Yeah.
12:47Your $200 a month plan isn't enough for Fable, allegedly. As usual, nothing is this simple.
12:53People seem to think what this indicates is that Anthropic wants to move Fable five to a higher tier. Maybe they'll do a thousand dollar a month sub later. Maybe they just wanna force you to spend more money.
13:04That's not usually what the case is here. If you think Anthropic's goal is to charge individual developers like you and me more and not to get as much money out of enterprises as possible, you don't understand them properly at all, much less the current state of the chip economy. Anthropic's problem here is that they have limited compute.
13:22They only have so many GPUs. They're getting a lot more from xAI, which is why they were able to do any of this in the first place.
13:29Fable would never have been in the sub tiers at all if it wasn't for the deal they cut to get access to Colossus from Elon. The three day testing window they had before was enough to know that a 100% access was not going to be realistic, especially when combined with the price drop. Remember, this model was originally set to be priced at a $100 per million tokens out, and it has been dropped to only 50.
13:51Sorry, it was more than that. I think it was $1.25 out. The price they launched with was less than half of the price that they had originally planned, likely because again, they had enough GPUs.
14:01The cost that we pay per token for most of these models is not even close to how cheap it is for Anthropic and OpenAI to run them. You can tell when you look at other open weight models and see how cheap those often are to run. These companies have massive margins because they also have massive expenses.
14:18Running the model is cheap. Making the model was expensive, and they have to charge accordingly. They also have limited availability of compute, and they don't wanna give the model out to other people with more compute unless they have like really, really crazy restrictions like they have with AWS, GCP, and now Azure.
14:33As such, they are charging based on a bunch of napkin math based on what is their availability, what is the demand they project, how much can they handle this over different loads, over different times of day, and how much abuse are those subscription users going to use. Because those sub users make up a very disproportionate percentage of Anthropic's compute usage relative to how much money they make for Anthropic.
14:55So why are they giving it back to us at all if they're just gonna cut it off in a week? A couple reasons. First and most obviously, the marketing hype.
15:03It gets people like me to use it more, to talk about it, and to be really excited, and then then go to our workplaces or tell our friends who have other workplaces that they should be using it too. Those companies have to pay the real rates or maybe like 10 to 20% discounts, and that's how Anthropic makes their real money.
15:18The other reason I think is much more interesting, they want to get a full week of usage from power users like us in order to figure out what pricing this should look like going forward and how much availability and how much, most importantly, GPUs they will need for this type of usage. By giving us a full week, they can see the ups and downs of how we use the model on weekdays and weekends, on and off work hours, all these other things that they need in order to know how much usage will happen so they can allocate the right number of GPUs.
15:45So this is a combination of a marketing experiment and user research in order to figure out how much they have to buy in order to do this right. And I have proof, by the way. Thorick posted the following.
15:56I've heard a lot of questions about Fable's availability on subscription plans. While it will come off of subscriptions after July 7, we aim to restore Fable as a standard part of our subscriptions as soon as capacity allows, as we mentioned in our original blog post. They want it in the subs.
16:11It can't be in the subs right now because they don't have enough capacity. As he clearly said here, as soon as capacity allows. This seven day window is a way for them to figure out how much we will use when there's a tiny bit of capacity available because it takes longer for the enterprises to ramp up.
16:28Once those enterprises have started to more heavily use Fable, Anthropix GPU availability is going to plummet. And letting us do the subs now is a combination of getting more information during this window, as well as taking advantage of the fact that that allocation has not been hit yet because the companies that are going to use it have not fully adopted it yet.
16:47This one week window is really interesting, and I think Anthropic, for the most part, is doing it the right way. But now we get to the problem many of us, honestly myself included, are facing. We can't use this in the sub, then the costs are going to be absurd.
17:01And even within the sub, the costs are a little intense and I've seen a lot of people hitting their limits way faster than they expected, making simple changes that I would argue the way they did it were simple errors.
17:11And a lot of these errors are absolutely understandable, which is why I wanna jump in front of as many of them as I possibly can right now. I'll keep it mostly brief here because the fun parts of my workflow are gonna be that separate video I mentioned.
17:21Definitely make sure you're subscribed and you hit that little bell button if you wanna see it because I'm gonna show you all of the tricks to maximize Fable for the little bit of time we have left before it leaves the sub tiers. But for now, I wanna help you guys reduce costs so you don't hit your limits quite as fast as you might otherwise.
17:35The first change you need to make, and just trust me on this one, I've pushed the hell out of this model. If you look at the effort selector and you're like, oh yeah, my work's really hard. I should probably use x higher.
17:45Maybe I really wanna use max because it's important to get every detail right. Or if you're really bold, you might go for ultra code and a super fancy animation here. I am going to highly recommend that you act as though nothing to the right of high exists.
18:01They do not meaningfully improve the quality of your work for the majority of work, and they do massively increase the usage. X high is pretty brutal. Max is just stupid.
18:12I've never seen Max give a better answer than high or x high, but I have seen it cost 10 to 50 times more. On high, my usage has been relatively tame. On my first day back with Fable, I got through about 25 PRs that were open.
18:26I closed most of them, rewrote some of them, updated and merged some of them, and I had Fable in one single thread handle all of that for me. It took about five hours maybe and it cost around 150 to $200 across the usage of both Fable and all the other things that I had it use as well, which is where we get to my other fun trick.
18:45Just because you have Fable selected doesn't mean Fable has to be the model doing the majority of the tokens. I wrote a couple tips on Twitter, the link will be in the description, but again, the next video is going to have way more detail about all of this. Just giving you a brief overview to help you save some money now.
18:59I already said this one using Fable on high effort is much much more reasonable and pretty much everyone I know using the model has either made the mistake and moved over to this or just started there and is happy. The rest here is what I wanna talk about now, which is that I taught Claude Code how to use codecs. I did this for a handful of reasons.
19:18One is because Codex usage limits are just absurdly generous, so tasks that I can route to that make sense to route to that. Second is that there's a lot of tasks that Claude is bad at that Codex happens to be pretty good at. Things like computer use that I find the Codex ecosystem, both the app and its integrations with Mac OS, as well as the model itself being better at vision and complex like recognition and manipulation of two d stuff, I found Five Five to be better overall at computer use, especially when used within Codex.
19:44But the third piece which is honestly kind of aligned with what I just said is that there's a lot of really token hungry tasks that I don't think Fable should be used for. Things that are super input token heavy like processing PDFs, auditing large code bases, scanning through large amounts of data and documents finding things or computer use because it has to take a ton of screenshots of the screen which become massive token hogs as you do more and more of them for long running tasks.
20:09All of these types of work are things that I like doing with Fable not necessarily things I like Fable doing, if that makes sense. Fable is so much better than other models at managing a fleet of sub agents that do different tasks and keeping everything moving forward, that with a little bit of effort with skill writing and some system prompt changes, you can get it to do really good stuff.
20:28If you wanna get to this now, the link's in the description to my tweet thread, but if you wanna wait, I'll have a very detailed description of how I did all of this coming super, super soon. These workflows got me to merge over 15 pull requests, many of which were massive, enclose way more stale ones, all in that $1.50 to $200 range, assuming I had paid full price for the tokens.
20:49But this all easily fit within my 200 sub tiers on Codex and Claude. I didn't even get close to maxing my weekly limit on the Codex sub. I did like 10 to 15% on it, I think, before hitting a reset.
21:00In the Claude code one, I got to like 40, but I also was very close to the reset on that anyways, just timing wise. So 40 to 50% utilization on the sub before reset, not too bad considering the amount of code I merged there. If I had a button at the top of every repo that I could click and have all the PRs get triaged, updated, merged, or closed, and it cost me $200, I'd press that button twice a day.
21:22It's so good. And as crazy as all of this might seem, I promise my workflow is actually not that complex. I don't have custom plugins.
21:30I don't have all these fancy MCP servers or anything. I'm just prompting in a way that works well with the model and the model seems to like. And the results are not a whole lot of spend and shipping a whole lot of code.
21:41So to recap super quick, you can reduce your cost massively by helping the model use sub agents on cheaper things, whether that is codecs or you're just telling it to use Sonnet or Opus for things that make sense. That helps a ton. Never ever touch x high or max reasoning with this model, or honestly, any anthropic model.
21:56It's basically a , and it's not worth pressing. The quality difference in the outputs is basically zero. I haven't seen any higher quality outputs for max, and I've only seen very small differences in x high.
22:08Just stick with high. Also, check out low and medium. You might be surprised how good those are too.
22:12Next, with the subscription availability, the thing they're doing right now is an experiment. It is inherently very limited.
22:18They need to know how much compute is available, how much the enterprises will use, and how much we will use before they can give more realistic estimates and get us the sub tier back the way we want it. And then we have the Nerf Performance, which I hope we all agree was absolute nonsense. And now the slate's cleared, I could do my video where I show all the fun things I'm doing with the model and the work flows I have built to maximize my usage of it.
22:39Everything from my instructions that allow Claude to call codecs for computer use and other tasks way more effectively, to my somewhat chaotic Vibe proxy that allows me to reroute across multiple different accounts without having to worry about it on the Claude code side. This has been working great across my entire fleet of machines, and I'm so thankful I set it up.
22:55Been way more useful than I expected. Hopefully you now understand the model's nowhere near as bad as Twitter seems to think. It is unbelievable using it every day, and I've never been so motivated to ship.
23:04I genuinely feel like I got a month of work done over the last three days, and I can't wait to build more, but I do have to go film one last video that I hope you guys like. So again, sub and hit that notification button because this is a really good week to be on top of your usage of these types of tools. I'm gonna go film that quick so I can get back to coding.
23:21So until next time, peace, nerds.
The Hook
The bait, then the rug-pull.
A viral tweet claims the newly-returned Fable model is broken, overpriced, and deliberately weakened. Theo, who says he doesn't even get along with Anthropic personally, spends the next twenty minutes showing the receipts that say otherwise.
Named ideas worth stealing.
06:23model
Two-stage constitutional classifier
- Stage 1: cheap probe reads internal model activations
- Stage 2: heavier classifier screens both input and output, only triggered when stage 1 flags risk
Anthropic's current safety architecture avoids paying full classifier cost on every request by gating the expensive check behind a cheap internal signal.
Steal forExplaining any tiered-cost safety/moderation system design
How they asked for the click.
VERBAL ASK
22:35next-video
“Sub and hit that notification button because this is a really good week to be on top of your usage of these types of tools.”
Soft, single-mention CTA folded into the sign-off rather than a dedicated pitch segment; paired with a teaser for a promised follow-up video on his actual workflow.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
