2026-10-07 · ← News
Anthropic deploys Haiku 5.5: Low-cost model now cuts cache read prices by 20%
Anthropic has released the Claude Haiku 5.5 model, aiming to lower costs for high-volume, routine, and speed-sensitive tasks. It costs around 75% less to run on average than Haiku 4.5 and also reduces cache prices for the Sonnet family.
The image could not be loaded.
Cheaper routine operations by 75%
Anthropic has unveiled Claude Haiku 5.5. The new iteration of its smallest model is tailored specifically for high-volume, routine operations, such as summarization, classification, database queries, and live support. Compared to Haiku 4.5, the new model runs around 75% cheaper on average.
For the first time, the company is also offering an adjustable effort setting for the model. Users can now choose the trade-off between intelligence and cost when using Haiku 5.5.
Agent subcontractor and cheaper memory
The new Haiku is not aimed at complex reasoning (that remains the domain of Opus 5.5 and Sonnet 5.5). Haiku 5.5 functions primarily as an agent subcontractor for rapidly handling sub-tasks in the background of coding and research workflows.
Alongside the launch of Haiku, Anthropic is halving the price of cache reads for the larger Claude Sonnet 5.5 model. According to the company, this equates to roughly a 20% drop in operating costs for agentic work on Sonnet.
Via API, not a universal discount
A price drop on the API doesn't necessarily mean a flat discount for end-user services. If you are a regular customer paying a fixed monthly subscription to Anthropic and not running agents via your own API credits, the cache discount won't show up on your bill.
Simultaneously, Anthropic is introducing a monthly API credit for Claude Max and Team subscribers. This is designed to support the creation of custom agentic applications on the Claude Platform.
Adoption in agentic workflows
The proof of Haiku 5.5's success won't be in broad intelligence comparisons with Opus, but rather in whether it catches on as the invisible engine in multi-agent applications built by developers and startups.
The speed and cost scalability for routine tasks will show just how sensitive the market really is to falling integration costs for adding AI to existing enterprise products.
Lilith's verdict
The core of the move is creating a mass factory floor: Haiku isn't here to think, but to rapidly ferry materials for the bigger, smarter machines.
I keep the external link at the end. First, a concise explanation here — no hunting across someone else's site.
Original source ↗ ↗