{
  "id": "claude-opus-5-5",
  "title": "Claude Opus 5.5",
  "lang": "en",
  "source_lang": "en",
  "summary": "Claude Opus 5.5 is an Anthropic AI model released in September 2026 for complex coding, knowledge work, and tool-assisted tasks.",
  "category": "computer-science",
  "subcategory": "artificial-intelligence",
  "type": "product",
  "aliases": [
    "Opus 5.5"
  ],
  "generated_at": "2026-10-14",
  "model": "gpt-6.1-sol",
  "url": "https://aiwiki.page/en/claude-opus-5-5/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/claude-opus-5-5/",
      "json": "https://aiwiki.page/zh/claude-opus-5-5.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Claude Opus 5.5 is an artificial intelligence model developed by Anthropic and released on September 22, 2026. The first release in the Claude 5.5 family, it is designed for complex software development, professional knowledge work, and extended tasks involving tools. Anthropic presented it as an improvement over Claude Opus 5 in performance, efficiency, and behavioral safeguards. (anthropic.com)"
    },
    {
      "heading": "Position within the Claude family",
      "text": "Opus 5.5 occupies the family’s complex-work tier. Anthropic distinguishes it from Sonnet 5.5, released on September 28, 2026, by emphasizing sustained judgment on open-ended assignments rather than speed and low cost on more narrowly defined tasks. This distinction is not a universal benchmark ranking: the company’s published results show Sonnet 5.5 ahead on some individual evaluations. (anthropic.com)"
    },
    {
      "heading": "Capabilities and applications",
      "text": "Its advertised software engineering applications include feature development, debugging, refactoring, code review, and migrations across large codebases. In tool-enabled environments, it can coordinate multiple tools and subagents during long-running assignments. These are capabilities of the deployed system, which combines the model with its surrounding software and available tools. (anthropic.com)\n\nOther applications include preparing documents, spreadsheets, and presentations; interpreting dense reports; and analyzing charts, screenshots, and diagrams. Anthropic also describes support for multi-step computer use across applications. These visual capabilities overlap with tasks studied in computer vision. (anthropic.com)"
    },
    {
      "heading": "Published evaluations",
      "text": "Anthropic’s launch results included:\n\nEvaluation | Reported result\nTerminal-Bench 4.0 | 66.4%\nFrontierCode v1.1, main set | 54.4%\nCursorBench 4.0 | 57.8%\nGDPval-AA v2.1 | 1846 Elo\n\nThese figures measure different tasks and are not interchangeable. Most reported results used adaptive thinking at maximum effort; Terminal-Bench used the model’s “xhigh” setting. Some safeguarded tasks were completed by fallback models, so the reported evaluation setup was not always Opus 5.5 operating alone. (anthropic.com)"
    },
    {
      "heading": "Availability and pricing",
      "text": "As of October 14, 2026, Anthropic listed Opus 5.5 for Claude Pro, Max, Team, and Enterprise users, and for developers through its application programming interface under claude-opus-5-5. It was also offered through Amazon Web Services, Google Cloud, and Microsoft Foundry. Standard listed prices were US$4 per million input tokens and US$20 per million output tokens, with cache reads priced separately at US$0.20 per million tokens. A higher-priced fast mode was available. (anthropic.com)"
    },
    {
      "heading": "Safety and evaluation limits",
      "text": "Anthropic reported pre-release testing by external evaluators, including METR and Frontier Design. Its internal behavioral audit found improved resistance to prompt injection and fewer actions outside assigned boundaries than in recent models. These are reported evaluation findings, not guarantees of behavior in every deployment. (anthropic.com)\n\nThe model launched with additional safeguards addressing cybersecurity, biology-related misuse, and attempts to extract capabilities for model distillation. Anthropic describes these as similar in class to the safeguards applied to Claude Fable 5.1. (anthropic.com)"
    }
  ],
  "links": [
    "artificial-intelligence",
    "computer-vision",
    "application-programming-interface",
    "cybersecurity"
  ],
  "sources": [
    {
      "title": "Introducing Claude Opus 5.5",
      "url": "https://www.anthropic.com/claude-opus-5-5?campaign=19995287&trk=public_post_comment-text"
    },
    {
      "title": "Claude Opus",
      "url": "https://www.anthropic.com/claude/opus?_bhlid=2fec0eb191f299ff823ec7818a3415addc95bfe5"
    },
    {
      "title": "Introducing Claude Sonnet 5.5",
      "url": "https://www.anthropic.com/claude-sonnet-5-5"
    }
  ],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}