LLMCostLab

PricesCompare → Claude Sonnet 5 vs Claude Haiku 4.5

Claude Sonnet 5 vs Claude Haiku 4.5: API cost compared

Claude Sonnet 5 lists at $2.000 in / $10.00 out per 1M tokens. Claude Haiku 4.5 lists at $1.000 in / $5.000. Sticker price only tells you so much — below is what each actually costs across eight production workloads.

advertisement
Cheaper on more workloads
Claude Haiku 4.5
wins 8 of 8
Blended rate gap
2.0x
at 3:1 in:out

Side by side

Claude Sonnet 5Claude Haiku 4.5
ProviderAnthropicAnthropic
Input per 1M$2.000$1.000
Output per 1M$10.00$5.000
Cached input per 1M$0.200$0.100
Context window200k200k
Blended per 1M (3:1)$4.000$2.000

Monthly cost by workload

Caching applied at each workload's typical hit rate. This is where the two models actually separate.

WorkloadClaude Sonnet 5Claude Haiku 4.5CheaperDifference
Customer support chatbot$257.80$128.90Claude Haiku 4.5$128.90 (50%)
RAG document Q&A$368.00$184.00Claude Haiku 4.5$184.00 (50%)
Coding agent$185.00$92.50Claude Haiku 4.5$92.50 (50%)
Bulk summarization$2,692$1,346Claude Haiku 4.5$1,346 (50%)
Structured data extraction$3,490$1,745Claude Haiku 4.5$1,745 (50%)
Long-form content generation$429.20$214.60Claude Haiku 4.5$214.60 (50%)
Email triage agent$805.00$402.50Claude Haiku 4.5$402.50 (50%)
Translation pipeline$2,810$1,405Claude Haiku 4.5$1,405 (50%)
advertisement

How to read this

Claude Haiku 4.5 is cheaper on the majority of these workloads, but "majority" is not the decision. The output:input price ratio is 5.0 for Claude Sonnet 5 and 5.0 for Claude Haiku 4.5. If your traffic is output-heavy — long-form drafting, translation — the model with the lower output price wins regardless of what the input price suggests. If you are input-heavy — RAG, bulk summarization, agents re-reading a codebase — the cached input rate matters more than either headline number.

Cost is also not the only axis. This site does not benchmark quality, latency, or rate limits, and a model that is 3x cheaper but needs two attempts per task is not cheaper. Use these numbers to size a bill, then validate on your own evals.

Run your own numbers

Cost calculator

Enter your own numbers, or start from a workload preset. Costs are monthly and update as you type.

Cheapest model
Cheapest → priciest spread
ModelIn /1MOut /1MMonthlyvs best