Two weeks into this series we've covered Cowork and Claude in Chrome, both about where Claude works. This week is about something more basic that a lot of business owners skip past without realising it's a real choice: which Claude model you're actually using.

Why this is worth five minutes of thought

Every time you open Claude, there's a model picker sitting at the top of the screen, and it's easy to leave it on whatever it defaulted to and never touch it again. That's a reasonable way to work most of the time, but the model you pick genuinely changes three things: how good the output is, how fast you get it, and, if you're paying by usage, what it costs. Using a heavyweight model for a quick one-line question is like hiring a consultant to check your spelling; using the lightest model for something that needs real judgement is how you end up rewriting Claude's output yourself anyway. Neither is wrong, exactly. They're just mismatched to the task.

The benefit: matching the model to the task means you're not paying more (in time or in cost) than the job actually requires, on either end.

The lineup, in plain English

Worth re-checking before we go further, because this space genuinely moves fast: even since this series was planned a few weeks ago, the lineup has shifted again. As of late September 2026, Anthropic's current model family, from lightest to heaviest, is Haiku, Sonnet, Opus, and Fable.

Haiku is the fast, lightweight option: everyday requests, quick answers, simple lookups.

Sonnet is the daily driver. Sonnet 5 is the default for most people's everyday chat, writing, and coding, and it's the model most business use lands on without needing to think about it further.

Opus (currently Opus 5.5) is built for deep research and reasoning that needs sustained thinking. It's a step up in quality, at a higher cost per use.

Fable (currently Fable 5.1) sits above that as the most capable tier, meant for long, complex, high-stakes work that can run with fewer check-ins along the way.

The benefit: you don't need to track every model release to use this well. Just knowing there's a genuine ladder, and that "the newest one" isn't automatically "the right one for this," is enough to make a better call than defaulting blindly.

What the "effort" setting actually controls

Alongside the model itself, you'll sometimes see an effort setting: low, medium, high, and a couple of levels above that. In plain terms, effort controls how hard Claude works on a given answer: higher effort means more thorough reasoning and generally better quality, at the cost of a slower, more expensive response; lower effort trades some of that depth for speed and economy. It's a second dial alongside the model choice, not a replacement for it. A lightweight model at high effort still won't out-think a heavier one, but a heavy model at low effort can be a reasonable way to get a fast, cheap answer from your best available model when the task doesn't need its full attention.

The benefit: on tasks you run often, this is a lever worth touching. Dropping the effort level on a repetitive, lower-stakes task can genuinely speed things up without meaningfully hurting the result.

A rule of thumb that covers most days

You don't need a flowchart for this. If the task is quick, high-volume, or low-stakes (drafting a routine email, summarising a document, answering a straightforward question), Haiku or Sonnet at their defaults will do it well and fast. If the task involves real judgement, a decision you'll act on, or research you need to trust (analysing a contract, planning something with several moving parts, working through a genuinely hard problem), that's when reaching for Opus or Fable earns its cost. Most day-to-day business use lives comfortably in the first category; it's worth noticing when you've actually landed in the second.

The benefit: this one rule catches the vast majority of cases without needing to memorise benchmark comparisons or keep up with every release.

Where to actually change it

In Claude.ai and Cowork, the model picker sits in the same place you type your message, and Claude remembers your last selection so you don't have to reset it every conversation. You can change the model, and the effort level, mid-conversation if a task turns out to need more (or less) than you expected. If you're on a Team or Enterprise plan, an admin can set an organisation-wide default model under Organization settings, either pinned to a specific model or set to automatically track Anthropic's current recommendation. That's useful if you'd rather not rely on everyone on the team making this call individually.

The benefit: for a solo user this is a two-second habit to build; for a team, it's a five-minute admin setting that means nobody has to think about it at all.

Bringing it together

None of this needs to become a chore. Left on its defaults, Claude will serve you reasonably well most of the time. This is really about knowing the dial exists for the handful of moments where it's worth turning: a report that actually matters going up a tier, or a routine task going down one to save time. Next week we're getting specific about one place this shows up constantly: Claude inside Excel and PowerPoint, and what a model-aware approach looks like there in practice.

If working out which model or plan actually fits how your business uses Claude sounds like more time than you have to spend on it, that's the kind of thing I help you with, so feel free to reach out.

Picking the Right Claude Model for the Job (Without Overthinking It)