Anthropic 发布 Claude Opus 5.5 模型:性能媲美 Fable 5.1,运行成本比 Opus 5 低 40%Anthropic releases Claude Opus 5.5 model: performance comparable to Fable 5.1, running cost 40% lower than Opus 5.


AI free token/AI免费积分
感谢IT之家网友 软媒使用者1296501、HH_KK、我家小妞很拽、乌蝇哥的左手、咩咩洋、非液化、华南吴彦祖、CuSO4_5H2O、xxx_x、Alita的迷弟、兰花是我 的线索投递!

IT之家 9 月 23 日消息,Anthropic 今日正式发布了 Claude Opus 5.5 模型,这是其全新 Claude 5.5 家族的首个模型,宣称在大多数工作上表现可媲美 Claude Fable 5.1,运行成本比 Opus 5 低 40%。

Anthropic 发布 Claude Opus 5.5 模型:性能媲美 Fable 5.1,运行成本比 Opus 5 低 40%

IT之家从 Anthropic 公告获悉,具体来看,Claude Opus 5.5 模型有以下提高:

表现维度,Opus 5.5 比 Opus 5 有了很大提高,官方宣传早期测试者在最复杂的工作中看到表现大幅提高。一名测试人员在不到一天内完成了 68 万行代码迁移 —— 这一任务本应耗时数周。Opus 5.5 擅长发现并修复软件中的低效程序 —— 当官方要求它将网页应用的每个页面缩短加载时间时,Opus 5.5 在 40 次尝试中成功了 39 次,而 Opus 5 仅仅做了一些小改进,也改变了应用的行为。另一位测试者让多个 Claude 模型依据单一提示构建游戏,Opus 5.5 凭借其画面和精致度取得了高于其他任何模型的评分。

安全维度,Opus 5.5 在 Anthropic 的自动化行为审计中取得了迄今为止所有模型中最高的分数,该套件在数千个模拟情形中测试 Claude。它比其他最近发布的模型采取难以逆转的行动或超出既定界限的概率更低,且比 Opus 5 更能抵抗被提示注入。官方还扩大了对齐测试范围,涵盖更长的任务、不可能完成的任务以及基于真实事件的情形。

由于 Opus 5.5 在生物学和网络安全领域与 Claude Mythos 5.1 相当,官方部署时采取了类似于 Claude Fable 5.1 的防护措施。经过审核的机构今天就可以申请生命科学验证项目,使用 Opus 5.5 开展生物学研究。将来几周,Anthropic 还将扩大对网络验证项目的访问,认证的网络安全从业者将可以使用 Opus 5.5 开展工作。

成本和速度维度,Opus 5.5 所需的计算量比 Opus 5 少,其定价也反映了这一点。官方测试表明,默认设置下,Opus 5.5 在典型工作负载下的成本比 Opus 5 低 40%。Opus 5.5 的输出速度也比 Opus 5 快 30% 以上。

Anthropic 发布 Claude Opus 5.5 模型:性能媲美 Fable 5.1,运行成本比 Opus 5 低 40%

除了降价外,Anthropic 还放宽了 Pro、Max、Team 等企业套餐的使用时间限制,还为订阅使用者给出速率限制重置能力,现在客户可以随时保存和使用。

沟通交流能力上,Opus 5.5 的沟通比之前的模型更自然。早期测试者发现其文字更清晰易懂,它把最关键的信息放在最前面,在长时间训练中成为更好的工作伙伴。正如一位早期测试者所说:“它的写作方式和我一样。”

Claude Sonnet 5.5 和 Claude Haiku 5.5 将在将来几周推出,表现、效率和安全性维度也有许多相同的改进。

在 Anthropic 官方的基准测试中,Claude Opus 5.5 在智能编码、计算机使用和知识工作维度保持领先。

Anthropic 发布 Claude Opus 5.5 模型:性能媲美 Fable 5.1,运行成本比 Opus 5 低 40%

本文来源:https://www.ithome.com/1/005/998.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

Thank you to IT Home users SoftMedia User 1296501, HH_KK, My Little Girl Is Very Arrogant, Fly Guy's Left Hand, Baa Baa Yang, Non-Liquefied, South China Wu Yanzu, CuSO4_5H2O, xxx_x, Alita's Fan, Orchid Is Me for their tip submissions!

IT Home, September 23 — Anthropic today officially released the Claude Opus 5.5 model, the first model in its new Claude 5.5 family, claiming performance comparable to Claude Fable 5.1 on most tasks, with running costs 40% lower than Opus 5.

Anthropic releases Claude Opus 5.5 model: performance comparable to Fable 5.1, running cost 40% lower than Opus 5.

According to IT Home's information from Anthropic's announcement, the Claude Opus 5.5 model brings the following improvements:

In terms of performance, Opus 5.5 represents a significant improvement over Opus 5. The official announcement states that early testers saw substantial improvements on the most complex tasks. One tester completed a 680,000-line code migration in less than a day — a task that should have taken weeks. Opus 5.5 excels at finding and fixing inefficient code in software — when asked to reduce the loading time of every page of a web app, Opus 5.5 succeeded in 39 out of 40 attempts, while Opus 5 only made minor improvements and also changed the app's behavior. Another tester had multiple Claude models build a game from a single prompt, and Opus 5.5 scored higher than any other model due to its visuals and polish.

In terms of safety, Opus 5.5 achieved the highest score of any model to date in Anthropic's automated behavior audit suite, which tests Claude across thousands of simulated scenarios. It is less likely than other recently released models to take hard-to-reverse actions or go beyond established boundaries, and is more resistant to prompt injection than Opus 5. Anthropic also expanded the alignment testing scope to include longer tasks, impossible tasks, and scenarios based on real events.

Since Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, Anthropic has deployed safeguards similar to those used for Claude Fable 5.1. Vetted institutions can apply today for the life sciences verification program to use Opus 5.5 for biology research. Over the coming weeks, Anthropic will also expand access to the cybersecurity verification program, allowing certified cybersecurity practitioners to use Opus 5.5 in their work.

In terms of cost and speed, Opus 5.5 requires less compute than Opus 5, and its pricing reflects this. Official tests show that under default settings, Opus 5.5 costs 40% less than Opus 5 on typical workloads. Opus 5.5 also outputs more than 30% faster than Opus 5.

Anthropic releases Claude Opus 5.5 model: performance comparable to Fable 5.1, running cost 40% lower than Opus 5.

In addition to price reductions, Anthropic has also relaxed usage time limits for Pro, Max, Team and other enterprise plans, and now offers rate limit reset capabilities for subscribed users, which customers can save and use at any time.

In terms of communication, Opus 5.5 communicates more naturally than previous models. Early testers found its writing clearer and easier to understand, it puts the most critical information first, and it becomes a better work partner during long training sessions. As one early tester said: "It writes the way I do."

Claude Sonnet 5.5 and Claude Haiku 5.5 will be released in the coming weeks, and will feature many of the same improvements in performance, efficiency, and safety.

In Anthropic's official benchmarks, Claude Opus 5.5 maintains its lead in intelligent coding, computer use, and knowledge work.

Anthropic releases Claude Opus 5.5 model: performance comparable to Fable 5.1, running cost 40% lower than Opus 5.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry