Claude Fable 5: Why I Still Choose Opus 4.8

Claude Fable 5 is back, but I'm sticking with Opus 4.8 for my daily work: 5 reasons why

Claude Fable 5 promises mythical power in AI, but unexpected restrictions make me doubt its utility for everyday use.

1. It's too volatile

I have moved beyond using AI to test AI tools. I now use Claude Code, Cowork, and Codex from OpenAI to accomplish real work.

Last weekend, I used a combination of Cowork and Codex to set up a critical protection solution to prevent a spam attack that was quickly turning into a denial-of-service attack. I worked tirelessly with these two AIs until I had a functional solution.

This is not the kind of work where you want your AI model to suddenly stop working. If I were relying on Fable and it abruptly ceased to function, that would be problematic. I prefer to use reliable tools, and so far, Fable has not earned that label.

2. The guardrails are shifting

According to a discussion thread on Reddit, it is unclear what triggers Fable's downgrade to Opus. This is unacceptable. While we know that AIs are non-deterministic and therefore not entirely predictable, it would be helpful to generally know when a project will be downgraded.

So far, Anthropic has not published clear definitions of the guardrails. My intuition is that they cannot, given that they are reacting strongly to the threat of a government ban. What might be approved today could very well be restricted tomorrow.

3. It's more expensive

Currently, until July 7, Fable 5 is included in your Max plan at no additional cost, but limited to 50% of your weekly usage limits. However, after this date, Fable 5 will no longer be included in subscription plans and will switch to a pay-as-you-go billing model, with standard rates of $10 per million input tokens and $50 per million output tokens.

I already pay $100 per month for Claude Max and use Opus 4.8. If Fable is likely to revert to what I already have in my subscription, it doesn't seem cost-effective to start using API billing for an unproven product. I prefer to wait.

4. It could be slow

I see complaints on Reddit reporting that Fable 5 is very slow. Some users compare it to running Ollama on a local server, which is indeed too slow. Others claim that Fable is not slower but has a different pace than Opus.

In any case, no one is reporting that Fable is extremely fast. I spend a lot of time waiting for Claude Code to do its job. Doubling or tripling its speed could justify additional expenses, but staying at the same pace is not enough to convince me to switch.

5. Opus 5 is likely on the horizon

Fable 5 and Mythos 5 are back and available for their respective user bases. Sonnet 5 was launched on July 1. Only Opus and the base model, Haiku, are awaiting updates to "5" versions. Anthropic hasn't told me anything about a release schedule for Opus 5, but it’s reasonable to think that an announcement will come soon.

On the other hand, Opus 4.8 is already very good. I actively use it, and it performs its role well. Yes, it has admitted to making a few mistakes, but who doesn’t? I’m not convinced I need an additional expense when my current solution works very well.

Conclusion

I'm not saying to avoid Fable 5. I don't know what you do with AI, the complexity of your projects, or your budget for cloud services. For my use as a solo programmer, columnist, and small business owner, I don't think I need it at the moment. But that's just me.

When Fable is integrated into the Max plan and replaces Opus or becomes a higher tier, I will likely give it a try. For now, I haven't encountered a situation requiring the additional capabilities that Fable is supposed to offer. Stay tuned; this story surely has more twists to come.