6 min read · updated 2026-10-14 · reviewed

Opus 5.5, Sonnet 5.5 or Haiku 4.5: pick a Claude model by cost per task

The cheapest model per token is rarely the cheapest model per finished task. A model that needs three attempts costs more than a stronger one that succeeds once. This guide gives a simple way to choose between Claude Opus 5.5, Sonnet 5.5 and Haiku 4.5, and to route different jobs to different models in Claude Code.

Editorial note

Written by Model.sale Engineering and reviewed against first-party documentation on 2026-10-14. Comparisons use the billing rules visible on the linked sources and do not claim universal model equivalence.

Measure cost per finished task

Pick five real tasks from your own backlog: a bug fix, a small feature, a refactor, a test-writing job and a code review. Run each on the candidate model, count how many attempts it needs to pass your tests, and read the charged cost of every attempt in the dashboard. Add the attempts together. That total, not the price per million tokens, is what the model costs you.

Where each model fits

Haiku 4.5 is the fast, low-cost option for small, well-specified work: renames, summaries, boilerplate, quick edits and commit messages. Sonnet 5.5 is the default for everyday coding because it handles multi-file changes at moderate cost. Opus 5.5 is for the hard cases: unclear requirements, deep refactors, architecture decisions and long runs where one wrong turn is expensive. Start at Sonnet and move a task up only when it fails twice.

Current rates

At the time of writing Model.sale charges $0.20 per million charged tokens for claude-haiku-4-5-20251001, $0.55 for claude-sonnet-5-5 and $1.00 for claude-opus-5-5, with cached input counted at 40%. The ratio matters more than the numbers: Opus costs about five times Haiku per charged token, so it only pays off when it saves several attempts or hours of your time. The live prices are on the Claude API page.

Route models in Claude Code

Claude Code lets you set a main model and a small helper model. Export ANTHROPIC_MODEL for the main model, for example claude-sonnet-5-5, and ANTHROPIC_DEFAULT_HAIKU_MODEL for helper tasks, for example claude-haiku-4-5-20251001. Inside a session, switch with the model command when a task needs more capability, then switch back. The setup guide has the exact environment variables.

Cap the downside

Create one API key per project or per teammate, give each a daily spend limit and a model allowlist, and keep expensive models off keys that run unattended. A prepaid wallet stops at zero, so a runaway loop cannot exceed the balance you deposited. Review the usage page weekly and move recurring cheap tasks to Haiku.

A simple decision rule

Use Haiku for work you could describe in one sentence, Sonnet for work you would give a capable colleague, and Opus for work you would normally do yourself. Re-test when a new model is released, because the right answer changes. Compare any alternative provider on the same five tasks rather than on a published price per token.

Sources and further reading

Checked on 2026-10-14. These links support the factual comparisons in this article; pricing and terms may change.

Use the numbers on your own workload

Open the cost calculator, choose a live model and enter a representative input/output mix before deciding.

Open cost calculator
← All guides