Amazon-backed (NASDAQ:AMZN) Anthropic has released Claude Sonnet 4.5, which it unabashedly refers to as “the best coding model in the world.”
Anthropic also said the new frontier model is the top model for building agents, using computers and has made substantial progress in reasoning and math over prior iterations.
Its software engineering capabilities earned a 77.2% accuracy rate through the SWE-bench verified benchmark. This outperformed OpenAI’s GPT-5 Codex at 74.5% and Google’s (GOOG)(GOOGL) Gemini 2.5 Pro at 67.2%. It also outperformed GPT-5 and Gemini 2.5 Pro in agentic terminal coding and agentic tool use. However, it scored slightly below these two models in high school math with no tools, graduate-level reasoning and visual reasoning. Claude Sonnet 4.5 outperformed GPT-5 and Gemini 2.5 Pro in financial analysis.
Anthropic also introduced some new features with the Claude Sonnet 4.5 release. This includes checkpoints in Claude Code, which allows users to save their progress and roll back to a previous state; a new context editing feature and memory tool for the Claude API; allows brought code execution and file creation in Claude apps; and made the Claude for Chrome extension for Max users.
“Claude Sonnet 4.5’s edit capabilities are exceptional – we went from a 9% error rate on Sonnet 4 to 0% on our internal code editing benchmark,” said Replit President Michele Catasta. “Higher tool success at lower cost is a major leap for agentic coding.”
Anthropic has made the Claude Agent SDK available to users, which provides developers the tools needed to build custom agents. What’s more, Anthropic introduced “Imagine with Claude” for its Max subscribers. This experimental preview creates “software on the fly” as it responds in real time to user requests to build programs.