Anthropic Unveils Claude Haiku 5.5: Speed Meets Affordability
Introduction to the New Model
Anthropic officially published a blog post yesterday, October 7, to announce the launch of its latest model. They proudly introduced Claude Haiku 5.5 to the artificial intelligence landscape. The company confidently describes it as the fastest, most economical, and most capable small model within the Haiku family to date. You can find more details about this release by visiting the official Anthropic Claude Haiku 5.5 announcement page.
Detailed Pricing Breakdown
Regarding the pricing structure, the provided blog post outlined the costs quite clearly. Here is the detailed pricing per one million tokens for the new models:
Haiku 5.5 (Prompts up to / over 100k)
- Cache Read: $0.01 / $0.05
- Cache Write: $0.125 / $0.625
- Input Token: $0.10 / $0.50
- Output Token: $0.50 / $0.25
Haiku 4.5 Pricing
- Cache Read: $0.10
- Cache Write: $1.25
- Input Token: $1.00
- Output Token: $5.00
Sonnet 5.5 Pricing
- Cache Read: $0.10
- Cache Write: $2.50
- Input Token: $2.00
- Output Token: $10.00
On average, running Haiku 5.5 costs approximately 75% less than operating Haiku 4.5. For requests that do not exceed 100,000 tokens, the input price stands at an impressive $0.10 per million tokens. Meanwhile, the output price is set at $0.50. However, once a request surpasses the 100,000-token threshold, the input and output prices increase to $0.50 and $2.50, respectively. These pricing structures are particularly relevant for developers evaluating cloud models for coding and market research, where managing token costs is a crucial factor in choosing the right tool.
Additional Pricing Adjustments and Performance
Simultaneously, Anthropic implemented a price reduction for Sonnet 5.5 cache reads. They slashed the price from $0.20 down to $0.10 per million tokens. Anthropic claims this strategic adjustment will lower the operational costs of Sonnet 5.5 by roughly 20% for most agentic tasks.
In rigorous knowledge work benchmarks, Haiku 5.5 achieved a remarkable score of 1620 on GDPval-AA v2.1. Furthermore, it secured a score of 1578 on AA-Briefcase v1.1. On the OSWorld 2.1 offline subset, the model attained an impressive 72.4%. Additionally, it scored 39.2% on the Terminal-Bench 4.0 evaluation.
New Adjustable Performance Features
Notably, Haiku 5.5 stands as the very first Haiku model to support adjustable performance settings. This innovative feature empowers users to perfectly configure the balance between cost and intelligence based on their specific needs. Moreover, Anthropic positions this new release as an ideal sub-agent model. It is designed to work efficiently alongside Opus 5.5 and Sonnet 5.5 when executing complex coding tasks.
