Anthropic Releases Claude Haiku 5.5 With 1M Token Context

What Happened

Anthropic released Claude Haiku 5.5 this week, a small-scale language model designed for high-volume tasks, offering a one-million-token context window and input pricing set at $0.10 per million tokens. The release positions the model as a cost-efficient option for developers running subagent pipelines and browser-based workloads at scale.

Key Specifications

Claude Haiku 5.5 carries three headline features: a one-million-token context window, adjustable inference effort, and its $0.10 per million input token price point. The adjustable effort setting allows developers to tune computational intensity against cost and latency requirements depending on the complexity of the task being processed. Output token pricing was not specified in the initial wire reports.

The one-million-token context window places Haiku 5.5 among a small group of commercially available models capable of processing very long documents, extended conversation histories, or large codebases within a single inference call. Google previously expanded context windows across its Gemini family as part of its 2024 model development cycle, a move that contributed to broader competitive pressure across the industry.

Background

Anthropic, founded in 2021 by former OpenAI researchers including Dario Amodei and Daniela Amodei, develops the Claude family of AI models and positions itself as a safety-focused AI research company. The Claude model line is structured across tiers, with Haiku representing the company's lightweight, faster, and lower-cost tier. Claude Sonnet and Claude Opus serve as mid-range and flagship tiers respectively.

Haiku 5.5 follows earlier releases in the Claude 5 generation. The version numbering indicates incremental updates within the broader fifth-generation Claude architecture. Anthropic has made prior Haiku models available through its API and through Amazon Web Services via the Bedrock platform, as well as through Google Cloud's Vertex AI.

The pricing of $0.10 per million input tokens is notably low for a model offering a one-million-token context window. For comparison, models with comparable context lengths from competing providers have historically carried significantly higher per-token costs, though pricing across the industry has declined steadily over the past two years.

Intended Use Cases

Anthropic has described Haiku 5.5 as suited for high-volume subagent tasks and browser workloads. Subagent architectures refer to multi-step AI pipelines in which one model orchestrates or delegates to smaller models running discrete sub-tasks. Browser workloads typically involve AI models parsing, summarising, or acting on web content in real time.

The combination of a large context window and low pricing is intended to make Haiku 5.5 practical for enterprise applications where many parallel inference calls are made simultaneously, and where cost-per-call is a primary deployment consideration.

Market Context

The release comes during a period of sustained competition among AI model providers on both capability and pricing. OpenAI, Google, Meta, and Mistral have each released or updated models in 2025, with pricing reductions and expanded context windows as recurring competitive levers. Anthropic's decision to pair a one-million-token context window with sub-dollar-per-million input pricing reflects that competitive dynamic.

Anthropic has not disclosed how many developers or enterprise customers are currently using the Haiku model tier, nor has it published benchmark comparison figures against competing small models in connection with this release.

What Happens Next

Anthropic has not announced a specific date for broader availability updates or additional Claude 5-series releases, but the company is expected to provide further technical documentation for Claude Haiku 5.5 through its developer portal in the coming weeks.