Anthropic has introduced Claude Haiku 5.5, marking it as the fastest and most affordable model in its lineup. Compared to Haiku 4.5, the average cost of running this new model has decreased by about 75%.

Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% less to run than Claude Haiku 4.5. https://t.co/jm06cZJfkV

— Claude (@claudeai) October 7, 2026

Focus on Cost-Effective AI Agents

The Claude Haiku 5.5 is primarily designed for high-volume and latency-sensitive tasks such as classification, summarization, query processing, database management, and launching sub-agents. Anthropic also markets the model for real-time user support, browser automation, and other scenarios where speed is crucial.

This model is the first in the Haiku line to offer a customizable “effort” level, allowing developers to choose between speed and reasoning quality based on specific tasks. It features a context window of 1 million tokens and a maximum response size of 128,000 tokens.

Moreover, Anthropic has significantly reduced pricing. For requests up to 100,000 tokens, one million input tokens costs $0.10, while output tokens are priced at $0.50. This represents approximately a 90% reduction in token costs compared to Haiku 4.5, although due to a new tokenization method, the actual savings in task execution average around 75%.

Haiku Demonstrates Enhanced Capabilities

Anthropic reports a substantial improvement across nearly all key performance metrics. For instance, in the OSWorld 2.1 computer management test, Haiku 5.5 achieved a score of 72.4%, compared to just 15.7% for Haiku 4.5. In the Terminal-Bench 4.0 for agent programming, the score increased from 0% to 39.2%.

Source: Anthropic.

However, larger models like Sonnet 5.5 and Opus 5.5 remain better suited for complex multi-step tasks. Anthropic suggests using Haiku 5.5 where scalability and cost-efficiency are more critical, such as when a primary agent divides work into numerous smaller tasks and delegates them to the more economical model.

Price Adjustments for Sonnet Model

The launch of Haiku 5.5 coincides with a revision of the pricing strategy for the entire product range. Anthropic has halved the cost of cache reads for Sonnet 5.5, reducing it from $0.20 to $0.10 per million tokens.

This adjustment is estimated to lower the operational costs of Sonnet 5.5 in most agent scenarios by about 20%. Additionally, Anthropic will introduce monthly API credits for subscribers of the Max and Team plans, enabling them to develop their own agents and applications based on Claude.

Claude Haiku 5.5 is available to developers via the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure. The model has also been integrated into Claude Code. Furthermore, Anthropic has added support for computer and browser management in beta versions of its Python and TypeScript SDKs.

As a reminder, Andreessen Horowitz recently assessed that top U.S. users spend an average of $903 per month on AI services.

Follow ForkLog on social media

Telegram (main channel) Facebook X If you spot an error in the text, highlight it and press CTRL+ENTER