Anthropic 推出了 Claude Opus 5.5 模型的使用指南，着重介绍了该模型在处理长任务、自主思考和代码编程方面的能力提升1。其中，多步工作能力被列为该模型在长任务执行中最主要的进展，包括在大型存储库中自动完成测试通过的功能1。
指南建议用户在优化提示词时移除"think carefully"等命令，因为 Opus 5.5 会在每次回复前自动进行思考1。此外，用户可以在模型运行时发送后续消息，无需重启进程1。针对工作流管理，指南推荐通过 CLAUDE.md 文件中的规则来指导模型何时停止询问、何时继续运行1。
对于大规模代码库的处理任务，用户可以要求 Opus 5.5 通过多个子代理并行分配工作，以支持审计、迁移或审查操作1。Opus 5.5 是首个推出时同时具备 Fable 级别生物和网络安全防护的 Opus 模型1。指南还提到 Fast mode 已作为研究预览版本发布，用户可通过 /fast 命令启用该功能1。
Anthropic has published guidance on effectively utilizing Claude Opus 5.5, its latest model iteration, highlighting significant improvements in extended task execution, autonomous reasoning, and code development capabilities 1.
The model demonstrates substantial advances in multi-step task handling, including the ability to automatically complete tests across large repositories 1. Users no longer need to include prompting phrases such as "think carefully," as Opus 5.5 now performs reasoning automatically before each response 1. The optimization guidance recommends leveraging a CLAUDE.md file containing rules to direct the model on when to halt requests and when to continue operating 1. For complex projects involving large codebases, developers can distribute work across multiple sub-agents operating in parallel to handle auditing, migration, or review tasks 1.
Workflow efficiency has been enhanced through the ability to send follow-up messages while the model is running, eliminating the need for session restarts 1. Opus 5.5 is the first Opus model released with Fable-level biological protections and network security safeguards 1. Additionally, a Fast mode research preview is now available, which can be activated using the /fast command 1.
评论
还没有评论，欢迎留下第一条。