Anthropic推出了Claude Opus 5.5大语言模型，声称其性能达到Fable 5.1水平，同时大幅降低了成本1。该模型的定价为每百万输入token 4美元、每百万输出token 20美元，相比前代Opus 5降低20%，缓存成本更是从0.50美元下降到0.20美元，整体成本下降约40%1。
在性能表现上，Claude Opus 5.5在多项基准测试中表现突出，在Artificial Analysis评分中领先1。该模型的改进方向包括写作质量、3D视觉理解、代理编码、安全性和自然交流能力1。用户反馈普遍积极，指出该模型比Opus 5更易于交互，代码质量与Fable 5.1相当但价格更低廉，且无需显式指令如"think hard"即可得到优质回复1。
Anthropic研究员Sam Bowman表示，"我们认为Opus 5.5的安全性明显高于前代，发布它很可能会降低不对齐相关的风险"1。尽管社区反馈整体正面，但在复杂推理和某些专业领域，该模型仍逊于Fable 5.11。
Anthropic has released Claude Opus 5.5, a large language model that achieves performance comparable to Opus 5.1 while substantially reducing operational expenses 1. The new model is priced at $4 per million input tokens and $20 per million output tokens, representing a 20% reduction compared to Opus 5, with cache pricing dropping 60% to $0.20 per million tokens, bringing overall cost savings to approximately 40% 1.
The model demonstrates improvements across multiple dimensions, including enhanced writing quality, three-dimensional visual understanding, agentic coding capabilities, and safety measures 1. In benchmarking assessments, Claude Opus 5.5 scored 58 points on Artificial Analysis ratings, outperforming comparable competitors across most tracked benchmarks 1. Sam Bowman from Anthropic stated that "we believe Opus 5.5 is substantially safer than its predecessor, and releasing it is likely to reduce risks related to misalignment" 1. Early user feedback indicates the model is more interactive than its predecessor, delivering code quality on par with Opus 5.1 while maintaining lower costs, and requiring no explicit performance-enhancing instructions such as "think hard" 1. However, the model still trails Opus 5.1 in complex reasoning tasks and performance within certain specialized domains 1.
评论
还没有评论，欢迎留下第一条。