Claude Haiku API Pricing
Claude Haiku costs $0.80 per million input tokens and $4.00 per million output tokens at list price. Here is the cost at volume and how it ranks against 6 other models (3 of 7 by blended price).
alternatives
Claude Haiku vs every other model on price
| model | blended / 1M | vs Claude Haiku |
|---|---|---|
| Gemini 1.5 FlashGoogle (Vertex) | $0.13 | 92% cheaper |
| GPT-4o miniOpenAI | $0.26 | 84% cheaper |
| Claude OpusAnthropic | $30.00 | 19× the price |
| Claude SonnetAnthropic | $6.00 | 3.8× the price |
| GPT-4oOpenAI | $4.38 | 2.7× the price |
| Gemini 1.5 ProGoogle (Vertex) | $2.19 | 1.4× the price |
cheaper than Claude Haiku · public list prices as of 2026-06 · estimates, not quotes
when_it_fits
When Claude Haiku is the right call
Anthropic's fastest, lowest-cost tier, aimed at high-volume and simpler tasks.
On blended price, Claude Haiku is mid-pack: 2 compared here are cheaper and 4 cost more. Whether its rate is worth paying comes down to fit. A model that resolves a task in fewer attempts or shorter prompts can come out ahead of a cheaper one that needs retries, so price it on your own evaluation set and token mix rather than the per-million rate alone.
faq
Questions & answers
- How much does Claude Haiku cost?
- Claude Haiku is $0.80 per million input tokens and $4.00 per million output tokens, at public list price (2026-06). At a typical 3:1 input-to-output mix that blends to about $1.60 per million tokens.
- What is a cheaper alternative to Claude Haiku?
- The cheapest option compared here is Gemini 1.5 Flash at about $0.13 per million blended, roughly 92% less than Claude Haiku. Whether it fits depends on your task: a cheaper model that needs retries or longer prompts can cost more in practice.
- How much does Claude Haiku cost to run an AI agent per month?
- On a representative agent workload (2,000 requests a day, 3 tool calls, RAG on), Claude Haiku comes to about $1,567 a month. That includes the agentic context tax, where every tool call re-sends the conversation and you pay for it again. Your real figure depends on volume and tokens, so run it in the calculator.
A model's list price is not its bill.
Prompt caching, context trimming, and the right tier per task usually move an LLM bill more than the model choice does. Book a call, or leave your email and I'll reach out.
Prefer proof first? See how this plays out in real case studies →