最强AI编码模型Claude 4来了，上线前竟试图勒索工程师，还想逃逸、反手举报欲做坏事的人类？

Core Insights - Anthropic has officially launched its next-generation Claude model, Claude 4, which includes two versions: Claude Opus 4 and Claude Sonnet 4, setting new performance benchmarks in code generation and advanced reasoning capabilities [1][2]. Model Performance - Claude Opus 4 is touted as the "world's strongest coding model," capable of autonomously running complex tasks for hours, while Claude Sonnet 4 shows significant improvements in precision compared to its predecessor, Sonnet 3.7 [1][2]. - Claude Opus 4 can run code refactoring tasks for up to 24 hours, while the previous models typically managed only 1 to 2 hours before errors increased [2]. - In benchmark tests, Claude Opus 4 achieved scores of 72.5% in SWE-bench and 43.2% in Terminal-bench, outperforming competitors [4][8]. User Feedback and Testing - Companies like Rakuten and Cursor have reported stable performance and advanced capabilities of Claude Opus 4 in high-demand tasks [4]. - Claude Sonnet 4 scored 72.7% in SWE-bench, surpassing Sonnet 3.7, and has been integrated as the underlying engine for GitHub's new Copilot model [7]. Model Features and Improvements - Claude 4 introduces a "memory" feature that allows the model to maintain external files for key information during long sessions, enhancing task continuity [9]. - The models also include a "thinking summary" feature for quick user reference and a dual-mode operation for rapid response and extended reasoning [10]. Pricing and Availability - The pricing structure remains the same, with Claude Opus 4 charging $15 per million tokens for input and $75 for output, while Claude Sonnet 4 charges $3 and $15 respectively [10]. - Both models are available through Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI, with Sonnet 4 accessible to free users and Opus 4 requiring a subscription [11]. Competitive Landscape - The launch of Claude 4 has intensified competition in the AI programming assistant space, particularly against OpenAI, which recently announced a $3 billion acquisition of AI startup Windsurf [19]. - Windsurf's CEO expressed dissatisfaction over the lack of immediate access to Claude 4 for their users, highlighting the competitive dynamics in the AI tools market [19][20].