LLM Pricing

← All Models

Anthropic: Claude 3.5 Haiku

🇺🇸 Anthropic · Claude 3.5

Input Price $0.800 per million tokens NT$25.6
Output Price $4.00 per million tokens NT$128
Context Window 200K tokens Output limit: 8K
OpenRouter Route Price Please verify with official pricing pages
Use this model via OpenRouter →

Overview

Anthropic: Claude 3.5 Haiku is a large language model API from Anthropic, part of its Claude 3.5 model family. Priced at $0.800 per million input tokens and $4.00 per million output tokens, it occupies the mid-range, balancing capability against running cost. Output tokens cost about 5× as much as input, so prompt-heavy workloads run noticeably cheaper than generation-heavy ones. A large 200K-token context window (≈300 pages of text) lets it take in whole books, large codebases, or lengthy transcripts in a single call. Beyond plain text it also accepts Image input, so it can be applied to multimodal tasks rather than text alone. On Artificial Analysis's Intelligence Index it scores 9 (F grade), a useful proxy for its general reasoning strength relative to the other models tracked here. All prices on this page reflect OpenRouter's routed rates and are re-synced automatically every day; confirm against the provider's official pricing before committing to production.

Dimension Unit Price (USD) Price (TWD) Effective From
Input per 1M tokens $0.800 NT$25.6
Output per 1M tokens $4.00 NT$128
Cached Input per 1M tokens $0.080 NT$2.6
Cache Write per 1M tokens $1.00 NT$32.0
Web Search per search $0.010 NT$0.32

Provider
Anthropic
Model Family
Claude 3.5
Version String
anthropic/claude-3.5-haiku
Status
Active
Modality
Text, Image
Context Window
200,000 tokens
Output Limit
8,192 tokens

Index Metrics

Cross-domain capability indexes evaluated by Artificial Analysis — Artificial Analysis

Coding Index 16 D Measured: 2026-09-08
Intelligence Index 9 F Measured: 2026-09-08

Benchmark Scores

Data source: Artificial Analysis

AA-LCR 27.3% F Measured: 2026-09-08
GPQA Diamond 40.8% D Measured: 2026-09-08
HLE 3.6% D Measured: 2026-09-08
IFBench 42.8% D Measured: 2026-09-08
MMMU Pro 45.6% C Measured: 2026-09-08
Non-Hallucination 58.9% Measured: 2026-09-08
Omniscience Accuracy 13.2% Measured: 2026-09-08
Tau2 24.6% Measured: 2026-09-08
TerminalBench 2.3% Measured: 2026-09-08

Use Case Analysis

Model strengths, weaknesses, and ideal use cases based on benchmarks and official documentation

Strength

  • Fastest Response Times Fastest comparative latency among all Claude models. TTFT 0.36s. Highest throughput 52.54 tok/s. Official
  • Exceptional Cost Efficiency Priced at $1/$5 per MTok, 5x cheaper than Sonnet 4.5. Official

Best For

  • Real-time User-facing Apps Fastest response times make it ideal for chatbots, autocomplete, and sub-second latency apps. Official
  • High-Volume Low-Cost Tasks At $1/$5 pricing, ideal for content moderation, log classification, simple extraction at scale. Official

Weakness

  • Limited Reasoning for Complex Tasks Struggles with complex reasoning, multi-step logical problems, and tasks exceeding 150 lines of code. Official

Not Recommended

  • Complex Software Engineering For complex coding tasks, multi-file refactoring, or code exceeding 150 lines, use Sonnet or Opus. Official

Key Insights

Key data points from this page for quick reference and citation.

  • Anthropic: Claude 3.5 Haiku Input price: $0.8/M tokens
  • Anthropic: Claude 3.5 Haiku Output price: $4/M tokens
  • Context window: 200,000 tokens
  • Provider: Anthropic
  • Model family: Claude 3.5
  • Modalities: Text, Image
  • Data source: OpenRouter, updated daily