Anthropic debuts Sonnet Claude 4.5

Claude Sonnet 4.5, which Anthropic recently launched, is referred to as the "best coding model in the world" and exhibits superior performance on development benchmarks while keeping the same API cost as its predecessor. The specifics: Sonnet 4.5 improves computer usage by around 20% over Opus 4.1 and achieves SOTA results in real-world software development (SWE-bench confirmed). In contrast to GPT-5-Codex's 7+ hour sessions, Sonnet 4.5 demonstrated the ability to code autonomously for almost 30 hours, producing 11,000 lines of code. Claude Code checkpoints, memory and context editing in the API, and a Claude Agent SDK for agent development are among the new features that Anthropic released. Additionally, the business made "Imagine with Claude" available to Max users as a five-day research preview, demonstrating real-time software production. This summer, OAI's Codex somewhat overtook Claude Code, although Anthropic may regain its lead with the introduction of a new top coding paradigm and platform enhancements. Another incredible accomplishment is a 30-plus-hour agentic session, which suggests a future filled with long-horizon jobs that will reveal unimaginable new capabilities.