Haiku 5.5
claude-haiku-5-5*For prompts up to 100K tokens. Above 100K: $0.50 input and $2.50 output.
CLAUDE API · VERIFIED OCTOBER 8, 2026
Compare current list prices for the Claude 5.5 family. Prices are USD per one million tokens on the Claude Platform; provider-specific rates can differ.
*For prompts up to 100K tokens. Above 100K: $0.50 input and $2.50 output.
Cache read $0.10 · five-minute cache write $2.50.
Cache read $0.20 · five-minute cache write $5.00. Fast mode costs more.
Choose a model and enter monthly token volume. The estimate uses the displayed list rates and runs entirely in your browser.
A cheaper model may use a different number of tokens or need more retries. Test on your own workload.
For high-volume summarization, classification, routing, compaction and real-time experiences where speed and unit cost matter.
For well-scoped coding, agents and everyday production work that needs a stronger balance of capability, latency and price.
For complex, long-running work where judgment and reliability matter more than the lowest token cost.
10M input and 2M output tokens cost about $2 on the listed Haiku short-tier rates, before caching or batch discounts.
The same volume costs about $40 on Sonnet. Track retries and accepted patches because task cost matters more than price per token.
The same volume costs about $80 on Opus. Reserve the higher rate for tasks where stronger judgment reduces costly failures.
Multiply input and output token counts by their respective per-million-token rates, then add cache and provider-specific charges.
Haiku has the lowest displayed token rates. The cheapest model per completed task can differ when retries, output length and quality requirements are included.
It can reduce repeated-input cost when the workload meets the provider’s caching rules. Cache writes and reads have separate rates.
Do not assume so. Verify the rate card for the exact platform and region you deploy to.
Check the official provider page before launch and whenever a model, context tier or hosting route changes.
Include prompt length, caching, Batch API and monthly request volume.