Model selection

Claude isn't one model — it's a family. Haiku 4.5 is the fastest and cheapest: great for quick edits, lookups, and high-volume simple tasks. Sonnet 5 is the balanced default: the newest Sonnet-tier model, handling most coding, writing, and explaining work at near-Opus quality. Opus 4.8 is the deep reasoner: the right call for complex architecture decisions and subtle multi-file bugs. Fable 5 (released June 2026) is the most capable Claude available to the public — state-of-the-art on nearly every benchmark, built on the same research as Anthropic's frontier Mythos model. In Claude Desktop you switch from the model picker in the chat (or type /model); either way the conversation continues, no restart needed.

Claude is a family of models, not a single thing. They share the same training lineage and values but differ significantly in capability, speed, and cost — and knowing which to reach for is one of the skills that separates casual users from people who get consistently good results.

The rule of thumb is simple: match the brain to the difficulty of the task.

Haiku 4.5 is the sprinter. It’s optimized for throughput — fast turnaround, low cost per call. Reach for it when you know roughly what you want and you just need execution: rename this variable across the file, generate this boilerplate, look up this value. On tasks where you’d be happy with an autocomplete, Haiku is the right level of engagement.

Sonnet 5 is the default workhorse, and it earns that status. It’s the newest Sonnet-tier model, closing much of the gap to Opus-level reasoning while staying fast and affordable. The overwhelming majority of real coding tasks — explaining an unfamiliar codebase, writing a new function from a spec, fixing a bug you’ve already diagnosed, reviewing a PR — land squarely in Sonnet territory. If you’re not thinking about which model to use, Sonnet is almost certainly correct.

Opus 4.8 is the specialist you call in when Sonnet isn’t cracking it. It thinks longer before it answers, can hold more context, and reasons across more moving parts simultaneously. That makes it the right tool for genuinely hard problems: complex architecture decisions, subtle bugs that have stumped you after multiple attempts, tasks that require tracking a large codebase and drawing non-obvious conclusions across it.

Fable 5 (released June 2026) is the most capable model Anthropic has made publicly available. It’s built on the same underlying research as Claude Mythos — Anthropic’s frontier model developed through Project Glasswing — with safeguards that make it safe for general use. It’s state-of-the-art on software engineering, knowledge work, vision, and scientific reasoning. Reach for it when Opus still isn’t enough: the very hardest design problems, enormous multi-file reasoning chains, or tasks where correctness matters more than cost. Note that queries touching cybersecurity, biology, chemistry, or sensitive domains automatically route to Opus 4.8 — a built-in safety boundary, not a limitation to work around.

The most common mistake is using Sonnet for everything — fine for everyday work, but under-resourced for the hard stuff. The practical habit: start on Sonnet, reach up when you’re stuck, reach down to Haiku when you’re in a quick-iteration loop, and reserve Fable 5 for the genuinely hard problems where it earns the cost.