IT之家 · 智能时代 · 10/8/2026, 06:26:08
Anthropic Launches Claude Haiku 5.5: 75% Cost Cut, Adjustable Performance & Sub-Agent Role
Anthropic released Claude Haiku 5.5, reducing average operational costs by ~75% compared to its predecessor, with input/output prices starting at $0.10/$0.50 per million tokens. It is the first Haiku model to support adjustable performance settings and is explicitly positioned as a sub-agent for coding tasks alongside Opus and Sonnet.
SOURCE COVERAGEFull Original Report
Regarding pricing, IT Home cites the blog post to provide the following price details:
| Price per Million Tokens | Haiku 5.5 (Prompts ≤/> 100k) | Haiku 4.5 | Sonnet 5.5 |
|---|---|---|---|
| Cache Read | 0.05 (approx. ¥0.34 CNY) | $0.10 (approx. ¥0.67 CNY) | $0.10 (approx. ¥0.67 CNY) |
| Cache Write | 0.625 (approx. ¥4.2 CNY) | $1.25 (approx. ¥8.4 CNY) | $2.50 (approx. ¥16.8 CNY) |
| Input Tokens | 0.50 (approx. ¥3.4 CNY) | $1.00 (approx. ¥6.7 CNY) | $2.00 (approx. ¥13.4 CNY) |
| Output Tokens | 2.50 (approx. ¥16.8 CNY) | $5.00 (approx. ¥33.6 CNY) | $10.00 (approx. ¥67.1 CNY) |
The average operating cost of Haiku 5.5 is approximately 75% lower than that of Haiku 4.5. For requests not exceeding 100,000 tokens, the input price is 0.50 per million tokens (approx. ¥3.4 CNY). For requests exceeding 100,000 tokens, the input and output prices are 2.50 (approx. ¥16.8 CNY) respectively.
Anthropic has simultaneously reduced the cache read price for Sonnet 5.5. This price has dropped from 0.10 per million tokens (approx. ¥0.67 CNY). Anthropic states that this adjustment can reduce the operating costs of Sonnet 5.5 by approximately 20% in most agentic tasks.
In knowledge work benchmarks, Haiku 5.5 scored 1620 on GDPval-AA v2.1 and 1578 on AA-Briefcase v1.1. It achieved a score of 72.4% on the OSWorld 2.1 offline subset and 39.2% on Terminal-Bench 4.0.
Haiku 5.5 is the first Haiku model to support adjustable performance settings, allowing users to configure the trade-off between cost and intelligence. Anthropic also positions it as a sub-agent model for Opus 5.5 and Sonnet 5.5 when executing coding tasks.
