Anthropic 推出 Claude Sonnet 5.5：更快 30%、單任務成本最多省 30%
Introducing Claude Sonnet 5.5
摘要分析
Anthropic 於 2026 年 9 月 28 日推出 Claude Sonnet 5.5，為 Claude 5.5 家族第二款，定位為 Opus 5.5 更快、更便宜的搭檔，適合日常任務、除錯與產出文件、簡報、試算表。公布成績含 Terminal-Bench 4.0 達 70.6%（Sonnet 5 僅 10.3%、Opus 5.5 為 66.4%）、OSWorld 2.1 電腦操作 80.1%、GDPval-AA 1844 Elo。定價維持每百萬 token 2／10 美元，快取讀取降至 0.10 美元，輸出速度快逾 30%，單任務成本最多省 30%。因資安能力達 Opus 5 水準，是首款內建資安防護的 Sonnet，已在 AWS、Google Cloud 與 Azure 上線。
- Terminal-Bench 70.6%，較 Sonnet 5 大幅躍升
- 定價不變，速度快 30%、單任務成本最多省 30%
- OSWorld 電腦操作 80.1%，接近 Opus 5.5
- 首款內建資安防護的 Sonnet
- 三大雲平台同步上線
Anthropic released Claude Sonnet 5.5 on September 28, 2026 as the second model in the Claude 5.5 family—a faster, lower-cost complement to Opus 5.5 aimed at well-scoped everyday tasks, bug fixing, and polished documents, slides and spreadsheets. Reported results include 70.6% on Terminal-Bench 4.0 (vs 10.3% for Sonnet 5 and 66.4% for Opus 5.5), 80.1% on OSWorld 2.1 computer use, 1844 Elo on GDPval-AA and 64.5% on Humanity's Last Exam with tools. Pricing stays at $2 per million input and $10 output tokens, cache reads drop to $0.10, output is more than 30% faster, and Anthropic says tasks can cost up to 30% less than Sonnet 5 because the model uses fewer tokens. It is the first Sonnet to ship with cyber safeguards, since its cyber capabilities match Opus 5; higher-risk cyber requests fall back to Sonnet 5. Available on AWS, Google Cloud and Azure with model ID claude-sonnet-5-5.
- Terminal-Bench 4.0 at 70.6%, a big jump over Sonnet 5
- Same price, 30%+ faster, up to 30% cheaper per task
- OSWorld computer use at 80.1%, near Opus 5.5
- First Sonnet shipped with cyber safeguards
- Available on AWS, Google Cloud and Azure
關鍵字：Claude SonnetAnthropic代理式程式開發電腦操作模型發布
來源原文：https://www.anthropic.com/claude-sonnet-5-5
本頁為報導內容的摘要與分析，版權屬原媒體所有；請以原文為準。