EN ▼
Favorites
My Favorites
View All
Market Cap Price 24h%

Disclaimer: Content does not constitute investment advice. Trading involves risks—please invest with caution!

Anthropic 发布 Claude Sonnet 5.5:编程能力超越 Opus 5.5,价格仅为其一半

2026-09-29 06:32:43
Bookmark

Anthropic 发布 Claude Sonnet 5.5:性能飞跃与成本优势并存

周一,Anthropic 正式发布了 Claude Sonnet 5.5,这是对今年6月推出的 Sonnet 5 的一次重大升级。Anthropic 表示,这款中级模型的速度比其前代产品快了 30% 以上。

“Sonnet 5.5 在日常范围明确的任务、修复代码漏洞以及制作精美的文档、演示文稿和电子表格方面表现最为出色。它还具备敏锐的设计眼光。” Anthropic 在官方声明中写道。

价格策略与综合性价比

Sonnet 5.5 的定价保持不变:输入令牌(input tokens)每百万个 2 美元,输出令牌每百万个 10 美元。令牌是 AI 读取和写入的文本片段,长度略短于一个单词,企业通常按百万为单位计费。这一价格是 Opus 5.5 的一半。然而,由于 Sonnet 5.5 在完成每项任务时消耗的令牌量减少了近三分之一,因此其实际运行成本低于 Sonnet 5。

编程能力卓越

该模型在编程领域表现尤为突出。在 Terminal-Bench 4.0 测试中——该测试旨在评估 AI 代理能否通过自行键入命令来完成复杂的职业任务,得分以完成任务的比例计算——Sonnet 5.5 取得了 70.6% 的成绩。相比之下,Opus 5.5 得分为 66.4%,而上一代的 Sonnet 5 仅为 10.3%。

In short, this cheaper model does more work. Independent testing firm Artificial Analysis conducted its own tests and confirmed this: Sonnet 5.5 scored 63.6%, Opus 5.5 scored 59.6%, and OpenAI's GPT-6 Astra scored 59.1%.

Comparison of performance under different settings

Scores also depend on "effort setting," which is a regulator that allows the model to spend more time thinking to get better answers, but also leads to increased costs. Anthropic said that in high-effort mode, Sonnet 5.5 performed comparable to GPT-6 Sol in the FrontierCode benchmark, but the cost per task was only one-fifth that of the latter.

In the GDPval-AA benchmark test-which uses a system similar to the Chess Elo rating to rate real-world professional work across 44 professions-Sonnet 5.5 scored 1844 and Opus 5.5 scored 1846, almost even. GPT-6 Sol scored 1487.

Market competitive landscape

Competitors are also matching prices. Last week, OpenAI lowered the price of GPT-6 Sol to $2 and $10, and the mid-tier model GPT-5.6 Terra was priced at $2 and $12 respectively. Anthropic did not publish benchmark results for the Terra model.

Generating content and cost analysis

Sonnet 5.5 是一个“话痨”模型。在最大努力模式下,它每个测试任务生成了约 193,000 个令牌,这是 Artificial Analysis 测得的最高值,比 Opus 5.5 高出约 60%。这导致每项任务的成本达到 7.60 美元,比 Sonnet 5 高出约 50%,这与 Anthropic 声称的可节省高达 30% 成本的论点相矛盾。

Anthropic 的成本节约主要体现在较低的设置上:在其应用程序中默认的中等努力模式下,该公司表示 Sonnet 5.5 能以不到十分之一的成本超越 Sonnet 5 的最佳编程成绩。Artificial Analysis 认为高努力模式具有最高的性价比。对于日常用户而言,只要保持低设置,就能以极低的成本获得接近旗舰级的编程体验。

数据来源说明与未来展望

Anthropic 的数据表为自我报告数据,而 Artificial Analysis 测试的是一个预发布版本,其中存在一个 Bug,Anthropic 预计该 Bug 对结果影响很小或略微低估了其分数。Anthropic 指出,在处理需要持续判断力的复杂工作时,Opus 5.5 仍然明显更强。

此外,专为高容量、对成本敏感的应用程序设计的 Claude Haiku 5.5 将在未来几周内推出。

Disclaimer:

All content published on this website, including hyperlinks, related applications, forums, blogs, and other media accounts, originates from third-party platforms and their users. CoinMarketInsight makes no representations or warranties of any kind regarding the website or its content. All blockchain-related data and materials are provided for informational and research purposes only and do not constitute financial, legal, or investment advice. Users and third parties are solely responsible for the content they publish. CoinMarketInsight shall not be liable for any losses arising from the use of this website. You should exercise caution and conduct your own independent research, review, analysis, and verification before making any decisions.

Read Full Article
More News
TOP

TOP