Claude Model Selector

Claude Sonnet 4.6. It handles nearly every task with the right mix of speed, capability, and cost. Drop to Haiku 4.5 only when you are running high-volume, low-complexity tasks where per-call cost matters. Escalate to Opus 4.7 only when you have a specific hard-reasoning task that Sonnet is clearly struggling on.

A rough rule: 1 token is about 0.75 English words, so 1000 words is about 1333 tokens. For a quick chat, input is typically 200-600 tokens and output 200-500. For long-document analysis, input can be tens of thousands. The calculator uses task-profile defaults that match real usage; adjust if your workload differs.

Prices shown here are Anthropic's public API pricing as of April 2026: Haiku 4.5 at $1/$5 per million input/output tokens, Sonnet 4.6 at $3/$15, Opus 4.7 at $5/$25. This page pulls from a central pricing data file that is verified weekly against platform.claude.com — verify against the source link above before committing to a production deployment.

If you are using Claude through the web or Claude Code, subscriptions are often cheaper. Claude Pro at $20/month unlocks Opus 4.7, Sonnet 4.6, and Haiku 4.5 access with generous daily limits. Claude Max at $100 (5x) or $200 (20x) is for heavy daily users. The API is better when you are building something programmatic, running high volume, or need fine-grained cost control.

Opus 4.7 is Anthropic's flagship reasoning model. At $5/$25 per MTok vs Sonnet 4.6 at $3/$15, Opus costs about 67% more than Sonnet. It runs more computation per token to deliver better results on hard tasks. For most tasks that extra capability is invisible — Sonnet would have produced an equally good answer. Use Opus when you can feel the difference.

Yes, and you should. A common pattern: use Haiku to triage or classify, then escalate only the hard cases to Sonnet or Opus. Claude Code does this automatically to a degree. If you are building a product, model routing is one of the biggest cost optimisations available.