Tosto Advisory
Home Meet Lucca Services Latest Articles Get in Touch Book a free call
Latest Articles · AI & Tools

When to use the different Claude models

With the release of Fable 5, Claude now counts 4 different models, while each model has 5 different levels of complexity to choose from (low, medium, high, extra and max), giving you a total of 20 options to choose from. And as if that was not overwhelming by itself, the models are constantly going through updates, like Opus releasing version 5 of the model.

By the end of this short article, I will break down how you can efficiently take advantage of all the current models, and their levels of effort.

Fable 5: the most powerful one

Fable 5 was launched in early June, and right after it got pulled out for a couple of weeks after the US government applied export controls to it, then came back on 1 July with tighter cybersecurity checks built in.

It also carries the highest price tag, roughly double what Opus 5 costs for a similar job. I'd suggest using it for the tasks that genuinely need the extra headroom, such as heavy coding or very complex research and development, while using the 'cheaper' models for everything else.

Sonnet 5: the medium tier has been working like a charm

Right off the bat, I will start by saying this has been the model I have been using the most. One of the reasons is, like most Claude users, I'm enrolled in the Pro plan ($20 instead of $100 for the Max) and usage is something I need to keep an eye on. Goes without saying that, for different tasks, you might need to increase the model in order to get more power, but I must say for 80% of the tasks I do, which include everything from simple coding to finance work and forecasting, Sonnet 5 is the real deal for a fraction of the usage cost of the big brothers (Opus and Fable).

Opus 5: the newest model

Opus 5 sits between the two. It launched on 24 July, priced the same as the last version, Opus 4.8, and it gets close to Fable 5 on the hard tasks for about half the price. It's now the default model on Claude Max and the strongest one on Claude Pro.

What sold me on it is that you can push it to Extra or Max for a genuinely hard problem and potentially get better results than Fable 5, or dial it back to Low or Medium for less demanding tasks, all for a fraction of Fable's price. The charts below, taken from Anthropic's own benchmarks, show the same pattern across three different tests.

Chart comparing Opus 5, Fable 5, Opus 4.8 and GPT-5.6 Sol on agentic coding score against cost per attempt, Frontier-Bench v0.1
Source: Anthropic, Frontier-Bench v0.1
Chart comparing Opus 5, Fable 5, Opus 4.8 and GPT-5.6 Sol on the Artificial Analysis Coding Agent Index against cost per task
Source: Anthropic, Artificial Analysis Coding Agent Index
Chart comparing Opus 5, Fable 5, Opus 4.8 and GPT-5.6 Sol on CursorBench score against cost per task
Source: Anthropic, CursorBench

In conclusion

First of all, I'd suggest playing around with the different models to see what works for you. However, if you want to stay efficient, I feel nothing beats Sonnet 5: working with complexity on medium and high will cover most of what you throw at it, on a daily basis. For everything else that requires more tech power, I'd go for Opus 5 on high effort (or above). And when you really can't get it right, or just feel Claude could do a better job, that's when I suggest reaching for Fable, as a last resort.

Curious how AI fits into a leaner operation?

Book a free 30-minute call. No pitch, just a conversation about where tools like this can and can't replace solid operations.

Book a free call