Anthropic · model

Claude 3 Haiku

$0.25/1M in · $1.25/1M out · 200K context. That's cheaper than 52% of the 298 models we track, by output price. Here's what it costs, how its price has moved, and where it fits.

output price · per 1M tokens
$1.25
input $0.25/1M · list
Context
200K
Input / 1M
$0.25
Output / 1M
$1.25
Modality
text + image
Provider
Anthropic
Tokenizer
Claude

Snapshot · source: provider list ↗ · how we measure →

cost at a typical workload
$200 / mo

For 1,500 input + 500 output tokens across 200,000 requests/month. That's 33× the cheapest tracked option (Ling-2.6-flash, $6.00/mo).

Tune the workload in the calculator →

output price history
output / 1M

Flat at $1.25 since tracking began 2026-06-14. Full history →

Where it fits

  • Multimodal prompts — accepts image input alongside text.

Watch for

    These notes are derived from price, context and modality — structural facts, not measured quality. Measured cost-to-finish-a-task lives on the real-cost index.

    Related models

    See how the leading models compare head-to-head →

    How to read Claude 3 Haiku's pricing

    Two numbers decide most of the bill: $0.25/1M for input (everything you send — prompt, context, attachments) and $1.25/1M for output (everything it generates, including hidden reasoning tokens). Output is priced 5.0× the input rate here, so the shape of your workload — how much it reads versus writes — matters as much as the headline figure. This is the provider's first-party list price, reconciled against the routed market each run.

    Frequently asked questions

    How much does Claude 3 Haiku cost per million tokens?

    Claude 3 Haiku is priced at $0.25 per 1M input tokens and $1.25 per 1M output tokens (first-party list price) as of the 2026-06-15 snapshot. Output is the figure that usually drives the bill. Enter your own token volumes in the cost calculator for a monthly estimate.

    What is Claude 3 Haiku's context window?

    Claude 3 Haiku has a 200K-token context window — roughly 300 pages of text. Prompt, attachments, conversation and the model's own output all share that budget, and every token you send is billed at the input rate.

    Related