Straight comparison of the three current Claude models by API price, so you can pick the cheapest one that's good enough for your task.
| Model | Input / 1M tok | Output / 1M tok | Best for |
|---|---|---|---|
| Claude Opus 4.8 | $5.00 | $25.00 | Hardest reasoning, long agentic tasks |
| Claude Sonnet 5 | $3.00 | $15.00 | Default for most coding/agent work |
| Claude Haiku 4.5 | $1.00 | $5.00 | High-volume, latency-sensitive, simple tasks |
Prices as of 2026-07. Always confirm current numbers at platform.claude.com/docs/en/about-claude/pricing before budgeting — pricing pages change.
A typical coding-agent turn — 2,000 input tokens, 500 output tokens, no caching — costs roughly:
| Model | Cost / request | Cost / 1,000 requests |
|---|---|---|
| Opus 4.8 | $0.0225 | $22.50 |
| Sonnet 5 | $0.0135 | $13.50 |
| Haiku 4.5 | $0.0055 | $5.50 |
Plug in your own token counts on the full calculator — it also models prompt-caching discounts (cache writes and reads), which can cut repeated-context costs by 80–90% on long-running agent sessions. See the caching math worked out in detail, or real monthly cost estimates for Claude Code specifically.
Start with Sonnet 5 as the default for coding and agent work — it's priced in the middle and handles almost everything well. Drop to Haiku 4.5 for high-volume, low-complexity calls (classification, extraction, simple chat) where latency and cost matter more than reasoning depth. Reach for Opus 4.8 only for the hardest reasoning tasks, since it costs roughly 4x Haiku's output price.