Fable 5 Is Back: Usage Limits, Opus 4.8 Fallback, and Real Costs
Prefer to watch? Watch on YouTube · Subscribe for new builds · Read the transcript
In this video I go over Anthropic's announcement that Fable 5 is coming back, and the conditions attached to it. It's a short one for anyone excited about Fable's return who wants to know what access and cost will actually look like before planning a big build around it. I recorded this in June 2026, and these tools and plans change fast, so check Anthropic's current docs for the latest details.
What I cover in this video
- 0:00: The catch: routine coding and debugging fall back to Opus 4.8, and Fable 5 is capped at 50% of weekly usage limits through July 7th
- 0:46: Why usage credits get expensive, and why open weight models are getting more attention
- 1:31: The 18 days without Fable after US export controls restricted access
- 2:18: GLM 5.2, OpenAI's unreleased models, and my take on Sonnet 5
- 3:04: How I plan to use my Fable 5 allowance on my ShipSpace app
- 4:36: The loop framework: a folder-based protocol for multi-agent coding
- 6:07: Why I want to test local LLMs and depend less on frontier models
The catch with Fable 5 coming back
Anthropic's post on redeploying Fable 5 has two details I think everyone should read before getting excited.
First, in the near term, routine tasks like coding and debugging will fall back to Opus 4.8. Coding and debugging are exactly what made Fable 5 so good in the 3 days we had it. If Opus 4.8 is handling those, I don't really see the point of having Fable back.
Second, Fable 5 is included for up to 50% of your weekly usage limits through July 7th. After that it's only available through usage credits. 50% might sound like a lot, but I'm pretty sure most people will hit it on their first day.
Some background: we went 18 days with no Fable. The US government applied export controls to the model, and Anthropic had to restrict access because of that.
Why the cost pushes people toward open weight models
This isn't getting any cheaper. If Fable 5 moves to usage credits, the $200 a month plan only goes so far. My guess is that businesses running Fable 5 in production could end up spending $10,000, $20,000, or $30,000 a month. I was expecting it to come back as part of the subscription, and it didn't.
That leaves room for open weight models. GLM 5.2 has been improving. From what I've seen, it's close to Sonnet 5 on benchmarks and much cheaper. I ran a few quick tests on Sonnet 5 in an earlier video. It did okay, but I found the release disappointing. If I'm paying through the API anyway, I'd rather use GLM 5.2 or one of their coding plans, which run around $80 a month.
OpenAI also has its Soul, Terra, and Luna models coming. They aren't public yet, but I expect them to be on the subscription.
Over time I want to test more local LLMs and see how they could automate some of the agentic workflows in my business, so I'm less dependent on frontier models. People like you and me keep getting pulled back and forth by whatever OpenAI and Anthropic decide. The claims that open weight models are dangerous or steal your data look like marketing to me, meant to keep you buying their product.
How I plan to spend my Fable 5 allowance
Since the allowance is limited, I'm going to use it on work that's worth it. I held off on some big features for my app, ShipSpace (a voice-controlled ADE), until Fable came back. Here's my list:
- Refactor a lot of the existing code and look for bugs
- Connect the Resend API so I get notified when people join our waitlist. That one is simple, so Opus can handle it.
- Build the loop feature, which is next on my workbench. I've already written the plan, and it's ready to wire up.
If I hit the 50% limit, I'll still feel it was worth it for a feature this big.
The loop framework
My friend Vamsi wrote the loop framework. He's been working on the architecture behind how agents operate and how LLMs build. It's a very simple protocol for multi-agent coding with no orchestration code: the folder structure is the control system. The idea is folder equals memory, prompt equals direction.
- The loop: plan, implement, and validate, repeating until all three agree.
- Three roles, three different models: a Planner (Claude) writes the plan, an Executor (Codex) writes the work in progress, and a Validator signs off on what moves into deliverables.
- One-way pipeline: nothing reaches deliverables until it's signed off, so the folder works as a state machine and you can see each phase's status at a glance.
- Simple coordination: there's one governing safety rule, and coordination happens through just two append-only files.
If you want to know more about how Claude and Codex compare as coding agents, see Claude Code vs Codex. My broader notes on multi-agent setups are on the AI Agents page.
Key takeaways
- Fable 5 is back, but in the near term routine coding and debugging fall back to Opus 4.8.
- Fable 5 is included for up to 50% of weekly usage limits through July 7th, then moves to usage credits, which will get expensive fast.
- Rising costs are pushing open weight models like GLM 5.2 forward as cheaper alternatives.
- Use your limited Fable 5 access on high-value work like refactoring, bug hunting, and big features, and hand simple tasks to Opus.
- Fable 5's return is temporary in the form we knew it, so ship your ideas fast.
FAQ
Does Fable 5 handle coding tasks after it comes back?
Anthropic says that in the near term, routine tasks like coding and debugging will fall back to Opus 4.8, even though coding was what made Fable 5 stand out.
What are the Fable 5 usage limits?
Fable 5 is included for up to 50% of your weekly usage limits through July 7th. After that it's only available through usage credits.
Why was Fable 5 unavailable for 18 days?
The US government applied export controls to Fable 5, which required Anthropic to restrict access to the model.
Are there cheaper alternatives to Fable 5?
Beau points to open weight models like GLM 5.2. He says it's close to Sonnet 5 on benchmarks, much cheaper, and offers coding plans for around $80 a month.
Transcript
Read the full transcript
0:00Fable 5 is back, but with a catch. Let's get into it. So, Anthropic just announced that Fable 5 is going to be back, but in the near term, right here, some routine tasks like coding and debugging will fall back to Opus 4.8. So, what's the point of Fable again if Opus 4.8 is going to be responsible for all the coding tasks? I'll put a link to this article down below. Another one you're going to want to pay attention to, you can see here that Fable 5 will be included for up to 50% of a weekly usage limits through July 7th, after which it will be available via usage credits. That's going to be costly. This is not getting any cheaper to play this game, which is why open weight models are coming up and moving ahead into the
0:46spotlight like GLM 5.2 and other local models are getting more notoriety because this is not getting cheaper. So, if you're going to be switching to usage credits for Fable 5, that $200 a month plan is only going to get you so far. And if you're going to be using Fable 5 in actual production, which many big businesses will be, you're going to be spending 10, 20, 30,000 dollars a month just to be able to run production code using Fable 5. So, that is interesting. I was expecting it to be released on the subscription, but no, it seems like until July 7th, we have up to 50% access, and then it will switch over to usage credits. So, again, Fable 5 is out, and it's coming out, and that's
1:31amazing, but with a catch. And I just wanted to make note of that. You know, 50% weekly usage limits through the 7th, I mean, come on now. That is that kills me, right? And then the one too is like, in the near term, routine tasks like coding and debugging, which is what made Fable 5 so good for the 3 days that we had it, will fall back to Opus 4.8. Make it make sense. Come on now. What is Anthropic doing? So, again, we've had 18 days with no Fable. The US government applied export controls to our newest model, Fable 5, and then it required us to restrict access. So, that is pretty sad. I just wanted to make this quick video for people that are getting hyped up, they can hardly sleep, can dream up
2:18ideas of what to build. And unfortunately, it's just opened the door for open weight models to take the lead. You know, GLM 5.2 has been cooking, as well as Open AI with their new Soul, Terra, and Luna models, which are yet to be released to the public, but I would expect those models to also be on the subscription. So, Anthropic, again, a very lackluster release with Sonnet 5. If you saw that video earlier that I released, I did a few, you know, quick little tests with Sonnet 5, and it performed okay, but at that point, GLM 5.2 also is very close as far as benchmarks, and about a tenth cheaper than Sonnet 5. So, if I'm using through the API, why don't I just use GLM 5.2 or get one of their coding plans for like
3:04$80 a month? So, again, I just wanted to make this just to keep you up to date, you know, I know it's exciting. Tomorrow's Fable Day, everyone's going to be excited, but at the same time, we really need to understand like what is the play here. You know, it was restricted for 18 days. What is the actual play? Like, what is their strategy? And I know, you know, 50% doesn't sound you know, it sounds like, "Wow, that's going to be a lot." You're going to hit that 50% usage limit, I guarantee you in the first day of using Fable 5. So, I'm going to be using Fable 5 to basically go into my app on Ship Space before I hit the 50% limit, and it just refactor a lot of the code, look for some bugs. I wanted to actually connect the recent API to be able to send me notifications of people who sign
3:50up on our waitlist. And I also wanted to implement this loop feature that I'm excited about. I held off the moment I realized that Fable 5 was coming today, and this loop feature is pretty exciting. I'm excited for integrating it into our Ship Space app. Basically, this loop feature is already in process. I have developed the plan for it, and getting it ready to actually write the code to wire it all up. So, basically, this loop feature is um the folder is the framework, right? And this was written by my friend Vamsi. He has been working on frameworks and actual architecture for building infrastructure, like how agents actually operate and how LLMs actually work, how
4:36they build. So, it's a dead simple protocol for multi-agent coding with no orchestration code. The folder structure itself is the control system. So, the thesis is folder equals memory, prompt equal direction. So, the mechanics is a loop, plan, implement, validate, repeating until all three agree. Three roles, three different models. Planner, Claude, writes the plan. Executor, Codex, writes the IR work in progress. Validator signs off into deliverables. Pipeline is strictly one-way. One governing safety rule, coordination is just two append-only files. Nothing reaches deliverables until signed off, so the folder is in a state machine. You read phase status at a glance. The big realization, the is literally your swarm
5:21milestone. So, this is our next in progress if you look at our workbench of things that I've been working on in this app, and basically, this is the next in line to be able to start building. And I just wanted to hold off until Fable, and if I hit that 50% limit, then at least it was worth it for building out this pretty big feature inside of our Ship Space app, and you can find that link down below if Do want a voice-controlled ADE. Shipping code at the speed of light? We'll put that link down below. You can sign up for the waitlist. Expected to get beta testers here in the next few weeks. But, yeah, kind of depressing. I wanted to make just note of, you know, the things that, hey, it may seem too good to be true. Well, it is. You know, it's going to be expensive. So, I hope you saved your money. I hope you, you know, put out a
6:07second mortgage on your house just to be able to get Fable 5 working. But, like I said, I'm excited for more of these open weight local LLMs to catch up and, you know, to start testing, you know, locally. Just to play around with it and to see how it could work inside of my business and how I can actually use local LLMs to, you know, automate some of these agentic workflows and rely less on frontier models. Cuz I feel like people who are like you and I are just being pulled back and forth by the whim of OpenAI and Anthropic and saying how open weight models are the devil and they steal your information and they're not good for humanity. You know, I think it's just a marketing tactic to get you to buy their product. So, Fable 5 was sorely missed and those are the two
6:52things that I'm going to be looking to do, wiring up Resend API to just get updates, which is pretty simple thing. I can have Opus do that. And, you know, have this next thing with the loop feature integrated inside of Ship Shape. And, you know, I could go through it all, but it's kind of lengthy. But, you know, at the same time at reading that, basically in the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. Like, what is Come on. And people are going crazy about it on Twitter. So, I don't know. Let me know your thoughts. Do you think that we got rug pulled on this? How do you feel about Sonic 5? I thought it was terrible. Like, really? Come on. That was pathetic. I did I just We're still in a weird era right now in the age of AI. I still feel like we're in the dark ages. We've entered the dark ages. We're in the so-to-speak crypto winter, AI
7:37winter of models just not getting much better as far as the frontier line, but you know, I'm excited for the future for people who want to have local hardware and run locally. So, I also opened up our community to a free tier. So, if you want to be around other builders, you can click the Shipping School link and we have a free community of 126 members and you can come chat with us and tell us what you're working on, get feedback on your app. We also have a paid upgrade where it gets you access to four calls per week as well as the classroom courses where we add four new courses every single week to the curriculum. But yeah, just wanted to make note of that, you guys. It's amazing that we're getting Fable 5 back, but just know that it's just temporary. So, ship your ideas
8:23fast and happy Fable day.
Ready to start building with AI?
Join Shipping Skool free — 4 live build sessions a week, courses, and a community that ships.
Join Shipping SkoolKeep reading
Fable 5 Review: Parallel Agents, Usage Limits, and Real Build Tests
I ran three Fable 5 agents in parallel at once. One redesigned my landing page in 16 minutes, one built a first-person RPG in 11, and one made a Remotion demo video in about 40. Fable 5 hallucinated less than Opus 4.8, but its usage limits make me unsure it's worth paying extra credits for.
Claude Opus 5 vs Fable 5: Cost, Speed, and 3 Real Build Tests
I ran Claude Opus 5 and Fable 5 on high effort through three builds: a Three.js MMO, a Remotion video, and a landing page rebuild. Opus 5 won the game and the UI test, the video was close to a tie, and Opus cost about half as much. It's my new daily driver, but Fable 5 is much faster.
Claude Code Just Dropped a Massive Update. Voice Mode, Loop Commands, and a Million Token Brain.
Anthropic shipped a dozen Claude Code updates in March. Voice mode, loop commands, a million token context window. Here is what actually changed my workflow.
