Claude Haiku 5.5
Anthropic · Low-cost, lightweight, Subtask agent
Claude Haiku 5.5 is Anthropic's low-cost, lightweight model for high-volume tasks and subtask agents. It supports text and image understanding with a 1024K context window, billed in USD per 1M tokens with prompt-length tiers (≤100K and >100K).
API pricing (official list price, USD per 1M tokens)
| Input (cache miss) | Prompt ≤100K $0.1 Prompt >100K $0.5 |
|---|---|
| Output | Prompt ≤100K $0.5 Prompt >100K $2.5 |
| Cache hit | $0.01 $0.05 |
| Tiered pricing | Billed by input length; the cost calculator selects the tier automatically |
| Billing currency | USD (US dollars) — the vendor's original billing currency, not converted |
Specifications
| Context window | 1024K tokens |
|---|---|
| Modalities | Text, Image understanding |
| Noted features | Low-cost, lightweight, Subtask agent |
| Vendor / channel | Anthropic |
| Price source | Anthropic official pricing |
| Verified | 2026-10-28 |
Site calculation: estimated monthly cost at default usage
Site calculation, not a vendor quote. Assumptions: 100 calls/day · 2000 input / 800 output tokens per call · 30-day month · 30% cache hit rate. Same formula as the cost calculator.
| Effective input price | 30% × $0.01 (cache hit) + 70% × $0.1 (standard) = 0.07 USD/1M tokens |
|---|---|
| Input cost | 100 × 30 × (2000 ÷ 1,000,000) × 0.07 = $0.44 |
| Output cost | 100 × 30 × (800 ÷ 1,000,000) × $0.5 = $1.20 |
| Estimated monthly cost | $1.64 |
Notes and limitations
· Prices are official published list prices per 1M tokens in the original billing currency (here CNY). Limited-time promotions, resource packs and subscription discounts are not included; the vendor's console prevails.
· "Verified" is the date this page's figures were last checked against the official source linked above. Pricing changes over time — confirm on the source page before making decisions.
· The monthly cost above is this site's own calculation under fixed usage assumptions. It is not a vendor quote and does not constitute purchase advice.
· This page records no benchmark results and claims no leadership on any task. Editor's note: this page records official pricing and specifications only and makes no quality or performance ranking claims. Evaluate output quality on your own workloads before committing.