5 Tricks to SAVE Your Claude Fable 5 Usage (80% Cheaper?!)

CChase AI
Computing/SoftwareSmall Business/StartupsManagement

Transcript

00:00:00What if I told you we could reduce Fable 5's cost by 80% while still beating Opus 4.8?
00:00:05Well, that is just one of five tricks I'm going to show you today that are all about reducing
00:00:10Fable 5's usage and token cost without losing what makes this model great. Because we all know
00:00:16the clock is ticking. We just got a few days left until Fable 5 is kicked off the pro and the max
00:00:21plan and we are stuck paying API prices. And on top of that, we're also usage capped. So it is
00:00:27imperative that we figure out quickly how to get the best bang for the buck with this new model.
00:00:34So that is exactly what we are going to talk about today. And let's dive in.
00:00:38So tip number one, this is the easiest one and arguably the highest leverage. It is changing
00:00:43the effort level. It is reducing the effort level with Claude Fable 5. Now by default, we are on high
00:00:50and some of you are crazy out there and I see you pushing it to extra high or even max. And the truth
00:00:56is you probably do not need that. First of all, what are we looking at here? Well, we're looking
00:01:01at a benchmark. This is deep sweet. This is one of my favorite benchmarks. It's all about long horizon,
00:01:06long running agentic tasks. And we see Fable 5 here, Opus 4.8, as well as GPT 5.5. Now I said in the
00:01:14intro, I said you could reduce costs by 80%. When we look at the max effort level, which by the way,
00:01:20doesn't give you that much of an upgrade in terms of, you know, pass percentage from extra high,
00:01:25it's costing us $22. Our average cost per task when looking at deep sweet. If we compare that to low
00:01:32is $3 and 76 cents, more than an 80% reduction in cost. Yet at low effort level with Claude Fable 5,
00:01:42we're doing better than max at Opus 4.8. So we're at 60% at Fable 5 low and we're at 59% with max Opus 4.8.
00:01:52And Opus in this case is $13 versus $3 and 76 cents. That's crazy. Like you could argue that's crazier
00:02:02than the jump from 59 to 70 is the fact we're doing this so much more efficiently. And when we pop up to
00:02:09medium, we're going up from 60% to 65% pass rate. And when we go up to high, we're at 69%. And on extra
00:02:16high, we're at 70% versus 59%. But at low, again, we're getting really solid outputs that are extremely
00:02:24cheap, even compared to GPT 5.5, which honestly is such a sleeper model. It's so good. I'm super excited
00:02:29to use 5.6 when it comes out. It's still doing better than medium, slightly more expensive, but not
00:02:35by a ton. Now we see this reflected in other benchmarks as well. Here's a look at frontier
00:02:39code accuracy versus cost. And this is coming from Anthropic itself. So in the orange, we have Fable
00:02:44in the green, we have Opus 4.8. And then down here at the bottom, we have 5.5. Look at low. All right,
00:02:50just a shade over $5. And the score is about 11%. If we look at Opus 4.8 on max, it's call it $11.
00:02:59And it's the same exact score percentage. So I'm getting the same pass rate as Opus 4.8 on max at
00:03:06half the cost. And if I go up to medium, I'm blowing Opus 4.8 out of the water while still
00:03:11being less expensive than extra high, which is where a lot of people sit for just default Opus
00:03:15settings. So with that in mind, does it make sense for us to sit on the default high level
00:03:19with Fable 5? I think when you look at both of these benchmarks, the answer is probably no.
00:03:24And we look at DeepSuite. DeepSuite is rather complicated tasks. Are you doing something very
00:03:30complicated? The less complicated of the task you're doing, if you're doing web design, you probably
00:03:35should be on medium or low. And that right away is going to reduce your costs and reduce your usage
00:03:41substantially. Substantially. So out of everything you see today, if nothing else, I want you to try
00:03:47doing a task on medium with Fable. Try doing a task on low and see how well it does. I think you'd
00:03:52be surprised and you're not going to get to that 50% of your weekly limit nearly as quickly as you would
00:03:57otherwise. And of course, to change the effort level, all you have to do is go into the terminal,
00:04:01do forward slash effort, and then set it where you want to. Now, before we jump into tip number
00:04:06two, quick word from today's sponsor, me. So I just released my Cloud Code Masterclass, and it is
00:04:11the number one way to go from zero to AI dev, especially if you don't come from a technical
00:04:15background. We focus on real use cases. It's updated every single week, and it also includes a
00:04:20Codex Masterclass and an agentic OS Masterclass. So if you're someone who's really trying to level up
00:04:26your AI game, you want to get serious about this, make sure to check it out. It's inside of Chase AI
00:04:30Plus. There's a link to that in the pinned comment. Now, tip number two, when it comes to reducing Fable's
00:04:35usage is also pretty straightforward. And that is stop using Fable to both plan everything and execute
00:04:42everything. Instead, make Fable the architect. Have it come up with the plan. And then depending on the
00:04:49complexity of the plan, have it divvy up the work to the appropriate model, whether that is Opus,
00:04:53whether that's Sonnet, or it's an outside model. It could be GPT 5.5. It could be something local.
00:05:00Fable's also smart enough to know which model is best for the job. And you can have Fable 5 explicitly
00:05:05call out those models in the plan. So Fable 5 does the plan, and then it says, hey, for the first part,
00:05:12I want Opus. For the second part, Sonnet makes sense. And for the third part, let's send that to
00:05:16OpenAI and bring in GPT 5.5. This is a perfect use case for something like the Codex plugin within
00:05:23Cloud Code. And if you haven't used that before, I highly suggest you do. You can totally bring in
00:05:27something like the Codex rescue function and have Fable call on that to give features to GPT 5.5,
00:05:34which again, awesome model. But if all that's too complicated and having it call out all these agents,
00:05:38you can do something as simple as simply throwing it in plan mode, having it come up with the plan,
00:05:44having it create some sort of markdown file, set the stage for your code base, and then simply
00:05:48spinning up another session with Opus and having it execute the plan that Fable laid out. You don't
00:05:53need to overcomplicate it, but that stops Fable from burning a bunch of tokens on low-level tasks that
00:05:58are going to be necessary for whatever you're creating. Now, tip number three is to bring in outside
00:06:02tools and skills like Ponytail that are all about reducing token count. Now, if you don't know what
00:06:07Ponytail is, I did a full video on it and its thing is like, Hey, Claude's pretty verbose. What if we
00:06:13gave it a set of guidelines to follow so that we still get the same outputs? It's still just as
00:06:18effective. It just writes less code to get there. Now, the thing with Ponytail is it gives us a bunch
00:06:23of benchmarks. The thing with the benchmarks are they've only been tested on Haiku 4.5 and Fable 5 is a
00:06:29different beast entirely. In my last video, I tested the numbers using Opus 4.8 and found that
00:06:35using Opus 4.8, these numbers were actually even better. It actually wrote less code. It consumed
00:06:41less tokens and it was faster. And so I went ahead and I ran some of the same benchmarks using Fable.
00:06:47So these numbers on the left right here that are underlined, this is the baseline. And then these
00:06:53over here on the right is what Fable 5 got using Ponytail. And this was on a medium setting. So
00:07:00across the board, essentially it put out less tokens. And in terms of cost, which is what we really care
00:07:06about because, you know, the tokens, like we do care if they're input versus output. At the end of the
00:07:11day, it was essentially 22% cheaper, which funny enough is actually even better than what they claim
00:07:17for Haiku. Now there's other skills like Caveman that claim to do the same thing. But the big picture
00:07:22with this tip is this is an expensive model. If there's stuff out there that can give us a 20% boost,
00:07:28it's worth experimenting with. So even if something like this is sort of like suspect to you,
00:07:32I think we shouldn't dismiss it out of hand because 20% is a lot of money when we're talking about
00:07:38thousands of dollars. Now on the surface, tip number four is the exact opposite of what I told you
00:07:43tip number two. Remember tip number two, I said, hey, Fable doesn't need to plan and execute,
00:07:48just have it planned. Well, in this tip I'm saying, let's not have a plan. Let's actually have Opus
00:07:53plan for Fable. Now, what I mean by this isn't every plan should be done by Opus and we have Fable
00:08:00actually create everything. I'm saying for a lot of our plans, they require research. And one of the
00:08:05best ways to research these days is with ultra code in dynamic workflows. Specifically, I'm talking
00:08:11about deep research. So for those of you who don't know, forward slash deep dash research is a built-in
00:08:17dynamic workflow you can use. And it's going to spawn a ton of sub agents. I use deep research in
00:08:22preparation for this video and it spawned 109 sub agents. First of all, would I want to run deep
00:08:28research with Fable 5 as every one of those sub agents? Absolutely not. I would blow through my
00:08:34limits. That makes no sense. But the real point here is I want to use a lower level model like Opus
00:08:39for deep research because research isn't something that requires some like super high in select reasoning
00:08:45level like Fable. However, Fable 5 doesn't necessarily have all the context of today. It's knowledge
00:08:51cutoff wasn't yesterday. We still need something to go out there on the web, gather information,
00:08:57do some baseline adversarial work to make sure that information even makes sense,
00:09:01and then hand that to Fable. And then Fable makes the plan, right? If we're going to plan something,
00:09:07we need information to start. And so I don't think it makes a lot of sense to have Fable go
00:09:12out and gather all the information. Let the lower level peons like Opus and Sonnet gather all that
00:09:18context and then hand it to Fable. And then Fable creates the plan and then they can hand it off.
00:09:24So Fable doesn't have to do everything in the planning stage. We can kind of give it a leg
00:09:29up, let it do the high level intellectual architecture work, and let these dumber models do everything
00:09:34else. And in that sense, dynamic workflows, ultra code, deep research is perfect for these low level
00:09:39models and saves that Fable usage for the more important things. Now, tip number five is something
00:09:44that actually came out a few months ago, and that is advisor mode. Advisor mode was originally shown
00:09:49with Opus and Sonnet working together. The idea is, this should sound familiar, is we have a smart model
00:09:56that is the advisor, that is the planner. It is handing off its plan to an executor, a lower level
00:10:03model, in this case, Sonnet. It is executing tools, it is reading, it is writing, but anytime it gets stuck,
00:10:11what does it do? Well, it shares its context with the advisor. The smarter model says, "Hey, here's what's
00:10:17going on. I'm stuck. What should I do?" And if this sounds like a more sophisticated version of
00:10:22everything we've been talking about up until this point, you would be correct. Now, Anthropic hasn't
00:10:26put out any official numbers of what this looks like with Fable as the advisor and having Opus be the
00:10:31executor or Sonnet, but we can make a few assumptions. What you see here is a graph from Opus and Sonnet 4.6.
00:10:40And this is one of the all-time graphs from Anthropic. I mean, just look at these axes,
00:10:43but what you got using advisor mode was a Sonnet that performed better for cheaper. So it was overall,
00:10:49it was just more effective. And you see that reflected here is well across multiple benchmarks.
00:10:54Now to actually use advisor in this way, you can't have your model set to Fable 5 because whatever
00:10:59model you have set, that is the model that is the executor. That's the model that's actually writing
00:11:04the code. So if I want Fable 5 as the advisor and I want Opus actually doing everything, then I need to
00:11:10make sure my model is set to Opus. Then I just need to do forward slash advisor. And then that's when
00:11:15you set the advisor model. So I do forward slash advisor Fable. Now Fable is the one that's going to
00:11:22be essentially telling Opus what to do. So if you're someone who really loves the idea of Fable purely
00:11:26acting as the architect, the conductor, and letting the lower level models do everything, this is
00:11:31definitely a path you should try out. So those are five quick tips for reducing your Fable 5 usage
00:11:36while still getting the most you can out of this amazing model. Hopefully Anthropic is nice to us
00:11:41and then just keep it on the pro and max plan. That would be great. And also, by the way, if you
00:11:45give us more than 50% of the weekly limit, that would be awesome too. But until then, we're going to
00:11:50work with what we have. So as always, let me know what you thought. Make sure to check out Chase AI
00:11:55Plus if you want to get your hands on my Cloud Code Masterclass. And besides that, I'll see you around.

Key Takeaway

You can dramatically extend your Claude Fable 5 usage and reduce costs by over 80% by lowering the effort setting, delegating execution to cheaper models, and using specialized prompting techniques like Advisor mode.

Highlights

  • Reducing Claude Fable 5 effort level from max to low cuts costs by over 80% while maintaining performance superior to Opus 4.8 at max effort.

  • Claude Fable 5 at a low effort setting achieves a 60% pass rate on DeepSuite benchmarks, outperforming Opus 4.8 at max effort which scores 59%.

  • Switching effort levels is done via the terminal command /effort.

  • Delegating low-level tasks to models like Opus or Sonnet and using Fable 5 only for architectural planning prevents excessive token burning.

  • Applying the Ponytail guideline set to Fable 5 reduces token count and results in a 22% cost reduction compared to baseline performance.

  • Using /advisor Fable while setting the active model to Opus allows Fable 5 to act as an intelligent architect directing a lower-cost executor model.

Timeline

Optimizing Effort Levels

  • Reducing the effort level is the highest leverage method for saving tokens and costs.
  • Low effort settings on Fable 5 outperform Max effort settings on Opus 4.8 in long-horizon agentic benchmarks.
  • Benchmarks show moving from max effort to low effort reduces average task cost from $22 to $3.76.

Defaulting to high or extra high effort levels is unnecessary for many tasks. Data from DeepSuite and frontier code accuracy benchmarks demonstrate that Fable 5 at a low effort setting achieves comparable or better pass rates than Opus 4.8 at max effort for a fraction of the cost. Users can adjust these settings in the terminal using the /effort command.

Strategic Delegation and Planning

  • Fable 5 should act as the architect rather than executing every low-level task.
  • Complex plans should be partitioned among different models like Sonnet or external models like GPT 5.5 based on suitability.
  • Using simple planning modes to generate blueprints allows lower-cost models to handle code execution.

Burning tokens by having Fable 5 both plan and execute all sub-tasks is inefficient. Instead, have Fable 5 design the architectural plan and then assign specific implementation steps to appropriate, lower-cost models. This approach keeps Fable's reasoning capacity focused on high-level orchestration.

Utilizing Token-Saving Tools

  • External guidelines like Ponytail reduce verbosity and token count.
  • Ponytail achieves a 22% cost reduction on Fable 5 by guiding the model to write code more efficiently.
  • Experimental tools designed to curb model verbosity should be tested despite their origins.

Techniques designed to limit model verbosity, such as Ponytail, are effective for Fable 5. Benchmarks show that following these specific prompting guidelines results in less code output while maintaining task effectiveness, leading to significant savings in token consumption.

Advanced Workflows and Advisor Mode

  • Use lower-level models for research-intensive tasks via tools like /deep-research before involving Fable 5.
  • Advisor mode enables a high-reasoning advisor model to direct a lower-cost executor model.
  • Setting the active model to Opus and the advisor to Fable 5 creates an efficient hierarchical workflow.

Instead of using Fable 5 for information gathering, use cheaper models for research tasks that require context but not high-level reasoning. Furthermore, /advisor mode allows Fable 5 to function as the conductor, providing high-level guidance to an executor model like Sonnet or Opus when it encounters difficulties. This structure optimizes for both cost and performance.

Community Posts

View all posts