Claude Haiku 5.5 Arrives: A Faster, Lower-Cost Model for Coding Subagents
Anthropic [launched Claude Haiku 5.5 on October 7](https://www.anthropic.com/claude-haiku-5-5), positioning it as its fastest and lowest-cost small model for high-volume work. The intended uses are specific: short summaries, context compaction, classification, database queries, quick edits, and subagent tasks within a larger coding workflow. The model is available on the Claude Platform as `claude-haiku-5-5` and through major cloud providers.
The useful developer question is not whether Haiku replaces a frontier coding model. It is whether a smaller model can handle the **many narrow steps** around a harder task at lower cost and latency.
The price depends on prompt length
For prompts **up to 100,000 tokens**, Anthropic lists **$0.10 per million input tokens** and **$0.50 per million output tokens**. Cache reads cost $0.01 and cache writes $0.125 per million tokens. For prompts **over 100,000 tokens**, the prices rise to **$0.50 input**, **$2.50 output**, **$0.05 cache reads**, and **$0.625 cache writes** per million tokens. That threshold matters for agents that repeatedly carry large repository context. [Source: Anthropic pricing table](https://www.anthropic.com/claude-haiku-5-5)
Anthropic says the model is around **75% cheaper per task on average** than Haiku 4.5. It also says token prices are 90% lower for prompts within the 100,000-token tier and 50% lower beyond it. The distinction is important: a lower token price does not by itself tell you the final bill if the new model uses a different number of tokens or requires retries. [Source: Anthropic announcement and pricing footnote](https://www.anthropic.com/claude-haiku-5-5)
Where it fits in coding
Anthropic suggests pairing Haiku 5.5 with Sonnet 5.5 or Opus 5.5 as a **subagent**. A larger model can own the overall change while Haiku handles a bounded search, a quick file edit, a summary of test output, or context compaction. Adjustable effort lets developers tune the cost-versus-capability trade-off for those steps. This is a workflow suggestion from Anthropic, not a guarantee that a particular task will be reliable without evaluation. [Source: Anthropic launch](https://www.anthropic.com/claude-haiku-5-5)
The company reports **39.2% on Terminal-Bench 4.0** and **46.4% on FrontierCode 1.1 Main** for Haiku 5.5. These are Anthropic’s published results under its own evaluation settings; sAIgnals has not run an independent benchmark. Anthropic itself says Sonnet and Opus remain stronger choices for complex agentic coding, whereas Haiku is better suited to narrower work such as compaction, summaries, and subagents. [Source: Anthropic performance and positioning](https://www.anthropic.com/claude-haiku-5-5)
Availability beyond the Claude API
Anthropic says Haiku 5.5 is available on its platform and through **Amazon Web Services, Google Cloud, and Microsoft Azure**. [GitHub also announced availability in Copilot](https://github.blog/changelog/2026-10-07-claude-haiku-5-5-in-github-copilot/) for Pro, Pro+, Max, Business, and Enterprise users, including VS Code, Copilot CLI, the cloud agent, and other supported clients. GitHub calls the model generally available but says the rollout is gradual; Business and Enterprise administrators can control it through model policies.
The announcement includes two other cost changes worth separating from Haiku’s own price. Anthropic cut **Sonnet 5.5 cache reads by half**, to $0.10 per million tokens, and announced monthly API credits for eligible Max and Team subscribers. Neither is a Haiku-only discount. [Source: Anthropic launch](https://www.anthropic.com/claude-haiku-5-5)
**Bottom line:** Haiku 5.5 is a promising candidate for repeatable, latency-sensitive steps in an AI coding system. Measure task success, retries, latency, and total cost on your own workload before moving a complex lead-agent role to the smaller model.