21:9 widescreen high-key lighting render: on the right, semi-transparent glass knowledge overlays extend along a light-band track toward the right, with warm cream and iridescent gradients of ice blue and coral pink; glowing etched glass text claude-5.5. This image serves as the hero for the Claude Opus 5.5 detail page.

Anthropic

Claude 5.5

claude-opus-5-5

The Claude 5.5 series includes three tiers—Opus, Sonnet, and Haiku—which can be chosen based on task complexity and cost, covering agentic coding, computer use, and high-concurrency batch processing

Route
claude-opus-5-5
Context window
1,000,000
Input
Max output
128,000 tokens
Knowledge cutoff
2026-06
Release date
2026-09-22
Streaming

Try claude-opus-5-5 online

Pricing

claude-opus-5-5 /1M tokens

Specifications Unit Price
Input $2.020
Cached input $0.100
Output $10.080
Cache write (5 min) $2.520
Cache write (1 hour) $4.032

Overview

Positioning Overview

claude-opus-5-5 is the text model in the Claude 5.5 series aimed at complex tasks, positioned for scenarios requiring multi-step judgment such as agentic coding, computer use, and knowledge work. It suits developers who need the model to autonomously orchestrate subtasks, handle long-flow code changes, or understand long documents, returning results as text through the unified HiAPI.ai interface. If the task is clearly scoped everyday bug fixing, batch summarization, or high-concurrency classification, the lighter tiers in the same series are usually more suitable.

Core Highlights

  • The highest tier in the series for complex judgment — compared with the lighter tiers in the same series, Opus 5.5 focuses on tasks requiring multi-step reasoning, such as agentic coding, computer use, and knowledge work.
  • Built for agent and computer use scenarios — suitable for multi-step processes such as orchestrating sub-agents, driving automated coding, and interactive operations, rather than calls that end after a single question-and-answer exchange.
  • Plain-text output, easy to store directly — returned uniformly as text that can go straight into downstream parsing, storage, or further processing without extra format adaptation.

Capability Description

Agent and multi-step task orchestration: can take on the reasoning role of an agent controller or sub-agent, breaking complex goals into multi-step operations, suitable for automated coding and interactive task flows.

Long-text processing for knowledge work: aimed at long-document understanding, writing, and organizing work, it can analyze and summarize materials and output directly usable text results.

Code understanding and modification: suitable for development tasks such as code review, bug localization, and fixing, with output delivered as textual explanations and code snippets.

Applicable Scenarios

  • Agent developers: controller reasoning for orchestrating sub-agents — input task goals and tool-call results to get next-step decisions and instruction text, used to drive automated coding or computer use flows.
  • R&D teams: code review and bug fixing — input code snippets and problem descriptions to produce problem-localization explanations and fix suggestions, integrated into code review or ticket-handling stages.
  • Knowledge workers: long-document understanding and writing — input long materials and writing requirements to produce structured analytical conclusions or finished text, used for deliverables such as reports and documents.

Limitations and Selection Advice

This model produces plain text only and does not cover multimodal input or output such as images. Capabilities and costs differ noticeably across tiers in the same series; before formal integration, confirm details such as context and rate limits for the actual sub-model. The structured data currently available on HiAPI.ai only indicates that the output type is text and does not provide context window or rate limit parameters. For clearly scoped tasks, high-concurrency or cost-sensitive batch processing, the lighter tiers in the same series are more suitable; when multimodal generation is needed, other models should be chosen instead.

  • New model New model launch: claude-opus-5-5
  • Parameter change claude-opus-5-5@aws parameters and pricing update

Related Models

View all models →

Same Type