When you open Claude you'll notice a model picker, and the names mean nothing until someone explains them. So here is the explanation, along with how I actually route my own work across them.
One caveat up front: model versions change quickly. This article describes the lineup as at August 2026, and I'll keep it updated, but always sanity check against Anthropic's own site if you are making decisions off version numbers.
The idea of tiers
Anthropic ships Claude as a ladder of models. Same family, same character, different trade offs between speed, cost, and raw capability. Think of it like an engine range across a vehicle lineup: the small engine is cheap to run and quick off the mark, the big one pulls the heavy loads. As at now, the ladder looks like this:
Claude Haiku (currently Haiku 4.5) is the fast, low cost model. It is genuinely smart, far beyond what small models were even a year or two ago, but its job is speed and volume. Near instant responses, tiny per task cost.
Claude Sonnet (currently Sonnet 4.6) is the workhorse in the middle, the best balance of intelligence, speed, and cost. It is the default for a reason, and for most people it is the right answer most of the time, including for serious coding and agent work.
Claude Opus (currently Opus 4.8) is the heavyweight of the main range, built for genuinely hard reasoning, gnarly debugging, and long complex tasks where quality matters more than cost.
Claude Fable (currently Fable 5) sits above Opus in a new top tier Anthropic calls Mythos class. It is their most intelligent generally available model, the one you bring in when the problem is at the frontier of what AI can do. There is also a restricted sibling called Mythos 5, the same underlying model offered to approved organisations with fewer of the dual use safeguards, which is not something most of us will ever touch.
How I route my own work
Haiku is for volume. Inside client automations, thousands of small decisions happen constantly: classify this email, extract the fields from this document, route this enquiry, tag this transcript. Each one is simple, and each one happens at scale. Haiku does these fast and for cents, and using a bigger model here is just burning money.
Sonnet is where I live. It is my default in Claude Code for building software, my default for drafting and analysis, my default full stop. The honest test of a workhorse model is whether you stop thinking about it, and I mostly have. If I have not deliberately chosen otherwise, I am on Sonnet.
Opus is for the hard yards. When Sonnet circles a nasty bug without landing it, when I am designing the architecture for a new system, or when a piece of strategy or writing really matters, I escalate to Opus and the extra depth is usually obvious in the first response.
Fable is for the frontier. The problems where I want the absolute best available thinking: thorny commercial strategy, complex multi system designs, the analysis I will be making real decisions from. It costs more, so the discipline is simple: use it when being more right is worth real money, which is less often than the hype suggests and more often than sceptics think.
The pattern underneath all of that: start cheap, escalate on evidence. Route each task to the least expensive model that clears your quality bar, and move up a rung when the output tells you to. Model maximalism, running everything on the biggest model, feels safe and quietly wastes a fortune at scale.
What this means if you just use the app
If you are chatting on claude.ai or the apps rather than building systems, the practical version is short: the default model is right for everyday use, and when you hit something genuinely hard, deliberately switch up a tier and ask again. Knowing the ladder exists, and actually using it, already puts you ahead of most users.
The model picker on claude.ai
Next up, the practical one: all the different ways you can actually use Claude, and how I split my own day across them.