Anthropic reinstated Fable 5 and shipped Sonnet 5 in the same week — a side-by-side test shows exactly which model to reach for and when the free window closes.
Fable 5 is back for a limited free window and is the best model for messy, high-stakes planning work, while Sonnet 5 (now the default) covers everyday tasks at a fraction of the cost and usage.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You use Claude regularly and want a practical rule for which model to pick for which task.
You run a small business or coaching practice and want a concrete prompt for turning scattered context into an actionable AI plan.
You want to know the free-access deadline and usage-limit mechanics before Fable 5 reverts to paid-only.
SKIP IF…
You already have a settled Fable/Sonnet/Opus usage routine and don't need the comparison reargued.
You're not a Claude user and have no interest in Anthropic's model lineup.
TL;DR
The full version, fast.
Anthropic's Fable 5 was pulled under US export controls and reinstated July 1, with free access ending July 7 or at 50% of weekly usage. In the same week, Sonnet 5 launched as the new default, closing much of the gap with Opus 4.8 at far lower cost. Testing an identical business-planning prompt in both models, Fable 5 produced sharper, more actionable output — picking a workflow that needed zero prep because the input already existed. The practical takeaway: use Fable 5 for complex, high-value planning while it's free, Sonnet 5 for everyday work, and Opus 4.8 only if usage cost isn't a concern.
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.
Fable 5 is back, Sonnet 5 launched, framing the which-should-you-use question.
01:01 – 04:09
02 · The facts: takedown, reinstatement, usage limits
Export-control timeline, July 7 free-access deadline, 50% weekly usage cap, token cost vs Opus 4.8, tighter safety filters, Sonnet 5 as new default.
04:09 – 07:55
03 · Turning chaos into a plan
Fable 5's strength at organizing messy business context; sets up the AI operating system demo prompt.
07:55 – 09:26
04 · Live demo: Fable vs Sonnet
Runs an identical 30-day AI productivity plan prompt in both models and compares outputs.
09:26 – 10:04
05 · Result comparison
Sonnet's suggestions are generic and require more manual prep; Fable recommends the workflow with zero prep because the input already exists.
10:04 – 11:04
06 · Grounding tip
Paste in a tool's own help docs (e.g. Claude Cowork) before asking Fable to plan around it — models aren't self-aware about third-party tools.
11:04 – 11:46
07 · Planning vs building apps
Fable 5 is strongest at planning/improving a tool, then building it; ranks planning above building.
11:46 – 12:12
08 · Recommendation summary
Use Fable 5 for messy/high-value work, Sonnet 5 for everyday tasks, Opus 4.8 if cost isn't a concern.
12:12 – 12:43
09 · Gemini Study Notebooks
Free adaptive AI tutor inside the Gemini app that quizzes and adjusts to the learner.
12:43 – 13:14
10 · Gemini Omni Flash
Conversational video editing, API-only for now; Igor is intrigued but not convinced it replaces real editing decisions.
13:14 – 14:02
11 · GPT-5.6 and sign-off
OpenAI's next frontier model announced for limited orgs; closes with a request for video ideas.
Atomic Insights
Lines worth screenshotting.
Fable 5's free access window closes July 7 or at 50% of weekly usage limits, whichever comes first.
Fable 5 consumes roughly 70% more tokens than Opus 4.8, not double as earlier reported.
Sonnet 5 replaced Sonnet 4.8 as Anthropic's default model for free and pro users, narrowing the gap with Opus significantly while using usage credits far slower.
Fable 5 has tighter safety filters than before; flagged prompts silently fall back to Opus 4.8.
When given an identical planning prompt, Fable 5 recommended starting with the workflow that needed zero additional input, while Sonnet 5's top suggestion required manual data entry first.
AI models are not self-aware about third-party tools by default — pasting in a product's help-center docs before asking it to plan around that tool meaningfully improves output quality.
Planning and improving an existing tool with Fable 5 outperforms using it to build the tool from scratch.
Takeaway
Pick the model by the job, not by the hype.
MODEL SELECTION
Reserve the most expensive, highest-capability model for messy, high-stakes planning, and default to the cheaper model for routine tasks.
When a flagship model is offered free for a limited window, that's the moment to run your hardest, messiest planning problems through it — not routine tasks you could do cheaper elsewhere.
A model that costs 70% more in tokens is only worth it when the task is complex enough that its output quality difference actually changes the outcome.
For a repeatable operational plan, look for the option that needs zero additional data prep over one that requires you to manually gather information first — that's the version you'll actually finish.
AI models don't automatically understand third-party tools well; pasting in a product's own documentation before asking for a plan meaningfully improves the specificity of the output.
When adopting AI planning tools for a small business, ask for explicit human-review checkpoints and time-saved estimates in the prompt itself, not as an afterthought.
Between planning, improving, and building with AI, planning and improving tend to produce the cleanest results — treat build-it-from-scratch as the lower-leverage use case.
Glossary
Terms worth knowing.
Export controls
US government restrictions that temporarily blocked public access to Fable 5, framed as the model being too capable for unrestricted release.
Grounding
Giving an AI model extra reference material (like a tool's documentation) so its answers reflect how that specific tool actually works, rather than a generic guess.
“Fable five, the best model ever available to the public that they took down, is back.”
clean cold-open hook, states the whole premise in one line→ TikTok hook↗ Tweet quote
09:26
“The recommendation here is completely different. It says build this first, the call converter, because the input already exists.”
concrete proof point of the Fable-vs-Sonnet difference→ IG reel cold open↗ Tweet quote
The Script
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.
17px
metaphoranalogy
00:00Fable five, the best model ever available to the public that they took down is back. And along with that, Sonnet five. And we still have Opus 4.8, was the best as of three days ago.
00:11How are normal people supposed to make sense of this? Well, I think they're not, at least not just based off the press releases. That's why in this video, I'm gonna break down what the differences are, what you should use, and I actually have some prompts here that I wanna show you that work particularly well with, uh, extremely powerful new generation model like Fable five.
00:28Not all use cases are created equal, and it's always fun to talk about things that work particularly well with a flagship release. And also the clock is ticking because this thing will not be free forever. It's literally just a few days.
00:39So I hope that this video will help you squeeze the most out of these model releases. Plus, we're gonna talk about other things that happened in the AI space briefly in this week's episode of AI News. Can use the show where I used to cover every single story, and now I focus more on the things that I think actually matter, like Fable five and Sonnet five this week.
00:55And as per usual, you always keep it practical. This is about the stuff that you can actually use. So let's get into it.
01:01And, yeah, the headline is simple. Fable five is back. Sonnet five was just released, and it's the new default inside of Claude.
01:09A lot of people immediately ask, okay. So which one is the smartest? I think a more practical question is, which one should I actually use in what scenario?
01:17So let's talk about that. But first, the facts. Fable, if you're not aware, was released a few weeks ago and then just after a few days, it has been taken down by the US government saying, hey.
01:26This is too powerful for the entire world to have access. So Anthropic just made the call to take it down for everyone. Now that those export controls were lifted on June 30 as of July 1, everybody has access to this again.
01:38Check it out. I go to claw.ai and there it is, Fable five. Now, you can immediately see that this is only available through July 7.
01:44So if you're watching this video shortly after the upload, you're in luck because you can still use this thing for free. One important nuance, it already shows here. This thing is gonna shut off when you hit 50% of your weekly usage limits.
01:55You can check those out under the profile, settings, and then here usage. Here is this weekly limits.
02:01They even have an extra Fable counter because they know everybody's gonna be all over this. If you're not familiar, Fable five eats your credit stuff fastest. You guys rightfully so corrected me on the last video that it's not exactly two x of Opus 4.8, which used to be the best model, you know, just as of a few days ago, but it's actually roughly 70% more tokens.
02:20We're gonna touch on usage and tokens in the end a bit more, but for now, just know that, yeah, this is the hungriest model, and you can't even use all of the usage you have in your account, just 50%. Additionally, they added more cybersecurity safeguards on Fable five.
02:33So some prompts that might be harmless might get flagged now, whereas before they would go through. It was already very strict. Now it's even tighter.
02:40Whenever it hits one of those safeguards, it's gonna fall back to Opus 4.8, their second smartest model. And then on a different note, they also launched Sonnet five, which is an upgrade from Sonnet 4.8 that we used to have before.
02:51If you're not familiar, it's basically the smaller brother of Opus. It's usually faster and way cheaper than Opus, but also not quite as smart. Now, in short, the gap between Opus 4.8 and Sonnet 4.8 just got smaller, way smaller.
03:05Opus is still better, but Sonnet is so much faster and more cost efficient. That's why they set it as the default. On many use cases, you won't even notice.
03:12As we ran all of our benchmarks, we basically found, hey, on some of them, it's a bit worse than Opus, but it's minimal and it's just a great deal. That's why they set it to the default for free and pro users. Also, it eats up your usage on your account way slower.
03:25Like, depending on what you're doing, it might be free forex, but that's just after one day of usage. But here's the important part. The important part is that Fable is still the GOAT.
03:33It's a whole new class of of model, and you should totally be using it for complex tasks, for things that need a lot of thinking. That's sort of the conclusion of this video that I'm giving you already, but I also wanna give you some things to do and some things to tackle right now before Fable five runs out and they switch token based usage where you'll literally be paying dollars for every single prompt that you run.
03:53As of now, you have a few days to kind of do these things that I'll tell you about. So let's do that. Okay.
03:58In no particular order, here are few use cases that I have found super, super useful. And for most everyday people that are not hardcore devs, this is what I would try now with Fable five because, again, it's just the best thing out there. First up, the high level idea here is turning chaos into a plan.
04:13So Fable five, more than any model before, is really, really good at taking in a bunch of chaotic documents, maybe launch documents or business context, and you can ask it to sequence and organize the work. What are the decisions that need to be made here?
04:26In what order should I tackle all of this madness? What am I missing? And what is the first step?
04:31It's so good at organizing chaos. It's better than anything I've seen before. This one would be hard to demo in the video and also you don't need a precise prompt.
04:38Kind of what I just said there roughly is gonna do it. It's more about the context you provide. And, no, you don't wanna overwhelm it.
04:44You don't wanna give it hundreds of pages, but if you have a lot of varied stuff and you're trying to make sense of it or trying to plan if that didn't work perfectly before, well, try that again in Fable. So the demo that I'm gonna do here for you is creating something like an AI operating system for your business or personal life.
04:59This has been an idea that I was pursuing early last year already, kind of building my own AI advantage OS and experiments, but people are doing this all over the place. They're building simple little hubs for themselves where different applications and prompts and loops all kind of live in one space. It's a beautiful thing, but many people are kind of overwhelmed by this idea and rightfully so.
05:19It can be kind of a lot at once, like, ego or AI operating system. What do I even do? Where do I even start?
05:24Well, I prepared a prompt here for you, and Fable happens to be the perfect partner for you to do that. So let me open up Claude here, and I'll put this prompt into the description below. And, really, this prompt is gonna have three blocks, so let's talk about these for a second.
05:36First up, it's context on the task and the task itself. So this is a demo example for you here, but let's say you're a nontechnical business owner that wants to turn their company into a simple AI operating system. And then the concrete task is gonna be building a thirty day AI productivity plan.
05:50But with this context here, it's gonna start designing tooling and not just telling you what to do, which is powerful, especially with Fable five. Okay. Here's really the key ingredient.
06:00These are things that I found to work really well and good ones to start with. First of all, we're gonna have the five highest value workflows to automate or assist with AI. Fable five more than ever is really good at being self conscious and realizing what it can do because it just trained on all the cybersecurity and development data.
06:16It's kinda my explanation for it, but really it's just better than any model before, so try that. The exact first workflow to build this week makes it very specific. Gives you, like, a single win.
06:25If you don't like it, by the way, can always iterate. What input each workflow needs. This is key because oftentimes, you don't need some sort of data or something from your side.
06:33What output it should produce, I mean, that's workflows one on one. Something Something comes in, something comes out. Make sure to define that here.
06:40The human review step, hey. If you're working with AI, you'll know that most use cases, especially the high leverage ones, have a human in the loop. So let's include that.
06:47The likely time safety tweak, this one is at the core of our teaching philosophy at the AI Advantage. We think about how we can save people time and then they can reuse that in other ways, which makes it feel like an investment rather than a chore. A simple tech stack using tools a normal business owner can handle, this just grounds it in the context we gave it in the beginning.
07:05And number eight, the first prompt I should paste into Claude to start building this. We don't want an abstract plan. We want, like, a concrete first step where we can just copy paste something and start making stuff.
07:14Nice. And then there's a bit more context, a small coaching and consulting business with free people, sell workshops, one on one coaching, and monthly membership. Anyway, there's more details here including the tools.
07:23This is the part that you would customize to yourself. And then I guess you could also customize this first piece. And you could go beyond business.
07:29You could make this personal. You could create this for a larger business. I just think this middle stack is really gonna be helpful to many people watching this video.
07:36So go ahead and try it. So I'm gonna run it in here. And you know what?
07:38For comparison, we're also gonna run it inside of Sonnet, which is great. Like, I'm not making this video to talk badly about Sonnet. It's a fantastic model.
07:46It's just Fable is better than anything else. Alright. Results are in.
07:49Let's start by reviewing the Sonnet output here. Again, identical prompt. Okay.
07:53So let's look at the top three, especially at the one that it highlighted as the first workflow to build and compare that just in the interest of time. So first of all, a client QA assistant, a chatbot with context, call to action plan converters who takes transcripts and tells you what to do next.
08:08Okay. Launch planner and email writer, which turns a launch idea into full email sequences. I can tell you this wouldn't be a valuable one, not with a ton of work that is.
08:17Now the table result. So the ideas are similar ish, but you can see they're all more specific, which always makes things more actionable.
08:25So for example, this one is called client q and a inbox assistant, and then it tells you that this one is specifically built for emails, and it actually considered a human review piece which Sonnet ignored. In this case, it would be reading, tweaking, and sending. And then we also have the time saved, which we asked for.
08:40Okay. Up until now, both pretty good actually. Small differences, but not night and day.
08:44But check this out. The recommendation here is completely different. It says build this first, the call converter, because the input already exists.
08:51Zoom generates the transcripts for you. There's zero preparation, and you can judge the output instantly because you were on the call. And then also the clients feel it.
08:59A same day recap with clear next steps looks premium. And all you need to do on this one is, well, just copy this prompt and as it says, paste in a transcript and boom, you have a first version of something working. Whereas I think the Sonnet suggestion was also good, but there's way more steps.
09:14You have to actually paste in the common questions. So when I told you that Fable is just more practical with these relatively complex tasks, I mean, is simple enough for demo here in the video, it just consistently holds true.
09:25I mean, the brief was getting a workflow set up immediately and this just gives you a prompt where you can just get a transcript and boom, you're up and running and there's no extra work. I like that. I know that's subtle, but those subtle differences are the difference between making things happen immediately or not.
09:38Now here's what I would actually do if I ran this. I would do the same prompt with Fable, but I would also attach details on my products, on my clients. And then if you have a free month for a twelve month goal, also include that.
09:50You could put that below this prompt or inside of the context. And then here's a bonus tip. If you go to Claude Cowork documentation and then you add in the support article on how to get started with Claude Cowork that includes some sub links to how to schedule tasks, or you add any other source or link that basically has context on how a product that you use works, in this case, Cowork, which is really good at intermediate agentic use cases, well, it will use that information, and this is important, to ground the truth that Fable five provides you.
10:20It knows a lot. It's very smart, but it's not all knowing. And oftentimes, you kind of have to nudge it a little bit and give it a chance to lure the good work by providing it with the context that matters, which in this case would be the help center of co work or maybe documentation on a app that you use.
10:35You might assume it has that, but usually it doesn't. AI models are notoriously bad at self awareness and knowing what exactly they can do. They're good, especially now with Fable, at making sense of what you need and what's doable with these apps and reconciling that.
10:50Whether it's this prompt or another, giving the grounding information and deep context on you and your business is a high level idea that is gonna apply now and two years from now. Alright. That was very use case y.
11:00Here's one more. Use it to plan your application or tool builds.
11:05So I know many viewers of this channel have probably built a version of a personalized tool, and Fable is just so good at planning them or improving them or building them too. But if I had to rank those three, planning and improving are kind of fighting for spot one and then building a spot two. Other tools are great at building too.
11:23It's just when you make the plan with Fable, gosh, everything just works better, looks cleaner. I guess you could extrapolate that to really planning anything that is a bit more complex. Fable five is just the best at that.
11:33But there you go. That rounds out our little, you know, Fable use case segment. So in conclusion on these three models, I would say use Fable five for the most valuable and the messiest work.
11:41Use Sonnet five for everyday work. You're gonna get a lot of mileage on your subscription. Or alternatively, use Opus 4.8 if you don't really care about the usage or the money it costs to buy these subscriptions that much.
11:51But we're probably gonna get Opus five soon, and that is kind of the overview of that. Alright. That was a bit more time than I planned, but I hope that was valuable to you.
11:59And if you wanna learn more about context management, there's a link in the description to the AI Advantage Club where I teach courses on context management just like this, but more detailed. It's a $1 trial for thirty days. You can check that out in the description.
12:09Let's get back to the video. A few more things that were interesting and hit my radar this week. First of all, Gemini study notebooks is an outstanding feature.
12:16It's a free adaptive AI tutor inside of the Gemini app. Basically, quizzes you and adjusts to you. I really love watching the educational space change, and I think this future where eventually everyone will have a personalized tutor is a very compelling one.
12:29And this is probably the most mature product that I've seen in that category, especially on a free tier. If you have kids and wanna use AI for tutoring, this is the product I'd recommend right now, and we're gonna be making a separate video just on this just because I like this so much. And then there's Gemini OmniFlash, kind of introduced a new thing that I haven't seen before, which is conversational video editing.
12:49It's in developer only access right now through the API. Eventually, this is surely gonna come to the Gemini consumer app. It looks rather gimmicky to be honest.
12:57You kind of talk to it and it does that, but that's not really how you edit videos. Editing is really about thousands of decisions and having a vision for the final product, which I guess you could communicate for voice. Eventually, again, I'm intrigued, not convinced yet.
13:11I'll keep my eyes on this as it matures. I wanna add that we're gonna get a new OpenAI model soon. They announced some GPT 5.6 models this week, but only certain organizations in The US have that for now.
13:21So we'll see, but that's gonna be the new frontier model family upgrading from GPT 5.5 now, which is also amazing, but probably at, like, Opus 4.8 or Sonnet five level.
13:31And, yeah, those are pretty much all the stories and use cases I wanted to show you this week. Leave me a comment on what you wish was in this video or what we covered on the channel. It's been a while since I asked for ideas.
13:40I always love hearing from you guys and, yeah, that's pretty much it. I hope this was helpful. My name is Igor, and I hope you have a wonderful week.
The Hook
The bait, then the rug-pull.
Anthropic's most powerful model was pulled from public access under a US export-control order — and it just came back, for a free window that closes in days. The video's real question isn't which model is smartest, but which one to actually reach for on a given task.
Frameworks
Named ideas worth stealing.
07:55list
The 30-day AI operating system prompt
Five highest-value workflows to automate or assist with AI
The exact first workflow to build this week
What input each workflow needs
What output it should produce
The human review step
The likely time saved each week
A simple tech stack using tools a normal business owner can handle
The first prompt to paste into Claude to start building this
An 8-part prompt template for turning a small business's scattered context into a concrete, buildable AI plan, tested against both Fable 5 and Sonnet 5.
Steal forany client-facing AI-adoption consulting engagement or internal ops-automation audit
CTA Breakdown
How they asked for the click.
VERBAL ASK
10:36product
“There's a link in the description to the AI Advantage Club where I teach courses on context management just like this, but more detailed. It's a $1 trial for thirty days.”
Soft mid-video mention, tied directly to the grounding-context tip just demonstrated rather than a hard sign-off pitch.
A 15-minute hands-on review of the most capable Claude model yet — what it does differently, what it costs, and what the pricing trajectory means for everyone using AI.
A hands-on tour of Claude Design 2.0 — shared usage pool, live canvas editing, brand-locked design systems, and a two-way Claude Code handoff — proven by building a landing page, pitch deck, and database-wired dashboard for a fictional fragrance brand in one sitting.
A 13-minute map of every Claude product, feature, and layer — from Chat to autonomous Scheduled Tasks — with one simple framework to keep it all straight.
00:01
00:14
00:34
00:42
00:53
01:04
01:18
01:29
01:38
01:49
01:57
02:11
02:20
02:32
02:40
02:48
03:12
03:21
03:37
03:41
03:53
04:05
04:17
04:31
04:35
04:46
04:55
05:05
05:24
05:29
05:53
06:08
06:13
06:20
06:30
06:41
06:56
07:08
07:14
07:28
07:48
07:58
08:09
08:16
08:31
08:44
08:51
09:02
09:16
09:22
09:46
09:50
10:03
10:13
10:35
10:43
10:55
11:15
11:36
11:51
12:02
12:07
12:17
12:38
12:49
13:14
13:21
13:44
13:56
15:10
12:43
11:14
12:47
15:30
26:28