Claude Haiku 5.5 Pricing
Anthropic ยท Live
Claude Haiku 5.5 costs $0.1 per million input tokens and $0.5 per million output tokens on the Anthropic API. Cached input reads are billed at $0.01 per million tokens. It supports a 1,000,000-token context window. Pricing verified October 8, 2026.
Claude Haiku 5.5 API pricing
Claude Haiku 5.5 API pricing is $0.1 per million input tokens and $0.5 per million output tokens, verified October 8, 2026.
- Input (per 1M tokens)
- $0.1
- Output (per 1M tokens)
- $0.5
- Cached input (per 1M tokens)
- $0.01
- Cache write, 5m TTL
- $0.13
- Cache write, 1h TTL
- $0.2
- Batch discount
- 50%
- Context window
- 1,000,000 tokens
Price history
See full Claude Haiku 5.5 price historyNo Claude Haiku 5.5 price changes have been recorded yet.
No price snapshots recorded yet.
Frequently asked questions
Short answers to the most common questions about Claude Haiku 5.5 pricing, limits, and capabilities.
Notes
ADDED 2026-10-08, read directly from platform.claude.com/docs/en/about-claude/pricing. Positioned for high-volume, latency-sensitive work: classification, extraction, routing. STANDARD: input $0.10 / output $0.50 / cache write 5m $0.125 / cache write 1h $0.20 / cache read $0.01 per MTok. Batch API halves input and output to $0.05 / $0.25. US regional endpoint bills at 1.1x. Context window 1M at standard pricing, stated explicitly on the page. PRICED BY PROMPT LENGTH โ A NEW DIMENSION. The $0.10 input rate applies to prompts UP TO 100,000 tokens. Prompts above 100,000 tokens pay a higher rate, which this schema has no field for. The stored figure is therefore the sub-100k rate and will understate cost for long-prompt workloads. Flag for Khaled: the calculator may need a long-prompt tier for this model. CACHE READ IS 10%, NOT 5%. The 5% cache-hit rate introduced with this generation applies to Opus 5.5 and Sonnet 5.5 ONLY. Haiku 5.5 keeps the older 10% convention: $0.01 is 10% of $0.10. Do not 'correct' this to 5%. VERSUS HAIKU 4.5: Haiku 4.5 is $1 / $5, so the published rates are 90% lower on both input and output. Anthropic's own launch post described Haiku 5.5 as 'on average 75% cheaper than Haiku 4.5'. The gap between 75% and 90% is unexplained from the pricing page alone โ it may be a blended figure across caching, batch or mixed prompt lengths, or it may account for the above-100k rate. Treat 90% as the published-rate comparison and do not cite 75% as if it came from the price table. Haiku 4.5 remains listed and is NOT deprecated.
