Claude Haiku 5.5 neural network is faster and 75% cheaper than the previous version

Anthropic has introduced Claude Haiku 5.5, the fastest and most affordable AI model in its current lineup. Compared to the previous generation, its cost has decreased by an average of 75%, and performance in some tests has increased several times over.

Usage cost

Claude Haiku 5.5 costs an average of 75% less than Haiku 4.5. For requests up to 100,000 tokens, the cost per 1 million input tokens is $0,10, and for output tokens, it is $0,50. This matches the base rates of the competing GPT-6 Luna model from OpenAI.

For requests exceeding 100,000 tokens, higher rates apply: $0,50 per 1 million input tokens and $2,50 per 1 million output tokens. At the same time, the new model is 90% cheaper than Haiku 4.5 for short requests and 50% cheaper for long ones.

Speed and specifications

Claude Haiku 5.5 supports a context window of up to 1 million tokens, allowing for the processing of large documents and long dialogues within a single request.

During preliminary testing, Asana recorded a reduction in task execution time of more than 30% and an acceleration in response generation of up to 2.5 times compared to the model previously in use. These results were obtained in specific work scenarios and do not imply similar acceleration across all tasks.

Testing results

In the OSWorld 2.1 benchmark, which evaluates an AI’s ability to control a computer and perform multi-step tasks, Claude Haiku 5.5 scored 72.4% on a limited set of offline tasks. For comparison, Haiku 4.5 scored 15.7%, GPT-6 Luna scored 48.9%, and Sonnet scored 5.5 – 83,9%.

In the GDPval-AA v2.1 test, which evaluates the performance of professional tasks, the new model scored 1620 points on the Elo rating system. The results for GPT-6 Luna and Haiku 4.5 were 1437 and 735 points, respectively, while Sonnet 5.5 received 1840 points.

In the Terminal-Bench 4.0 programming test, Haiku 5.5 achieved 39.2% compared to 0% for the previous generation. However, the larger Sonnet 5.5 and Opus 5.5 models remain more suitable for complex programming tasks.

New features and availability

Claude Haiku 5.5 is the first model in the Haiku family to offer adjustable compute effort. Developers can choose between deeper analysis for complex queries and faster execution for simple tasks at a lower cost.

The model is designed for automating customer support, processing and classifying data, generating summaries, and working with code. It can also be used as an auxiliary AI agent in conjunction with the larger Sonnet 5.5 and Opus 5.5 models.

Claude Haiku 5.5 is now available in the Claude web version, mobile apps for iOS and Android, Claude Code, and via API Anthropic under the identifier claude-haiku-5-5. The model is also supported by cloud platforms Amazon Web Services, Google Cloud, and Microsoft Azure.