CASE EVIDENCE / A RECORD
Replit 公开的代码审查与测试案例，来源为 官方页面，复核于 2026-06-26T20:20:00Z。
原始记录：Replit selects Claude 3.5 Sonnet as foundation for Agent 4, reaching 50M+ users without coding experience
Chinese Brief
Replit 公开的代码审查与测试案例，来源为 官方页面，复核于 2026-06-26T20:20:00Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。 Model Atlas 不把 benchmark、教程、发布说明或集合页包装成真实案例。
任务
这是一个围绕代码审查和测试生成的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：Replit, an agentic software creation platform, uses Claude to help anyone build and deploy software applications without coding experience. Michele Catasta, President and Head of AI, stated: 'We made the choice back whe…
公开材料提供原始证据链接和可访问产物，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：Replit reached 50 million+ users building applications without coding skills. Annual recurring revenue grew from $240M. Teams from Zillow to Duolingo deploy Agent 4 to ship internal tools. One user generated 36,000+ lin…
Claude 3.5 Sonnet 在该案例中承担代码审查和测试生成相关的生成、分析、编排或实现角色。 原始资料写作：Claude 3.5 Sonnet was the foundational model Replit chose to build their AI agent upon in 2024. The model's reasoning depth, vision capabilities for self-testing, and long-running task consistency enabled Agent 4 to man…
当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：Customer story is authored by Anthropic, with quotes from Replit's President. The platform now uses newer Sonnet/Opus versions for Agent 4, but the foundational choice was made with Sonnet 3.5.