Anthropic has unveiled its next-generation AI models, Claude Opus 4 and Claude Sonnet 4. This update is garnering attention particularly for its groundbreaking performance improvements in coding, as well as its ability to maintain consistent performance in complex, long-term tasks.
"World-Class" Coding Performance
Claude Opus 4 has demonstrated the best coding performance among currently released models, scoring 72.5% on SWE-bench. It also achieved 43.2% on Terminal-bench, showing that it can provide practical assistance with complex software development tasks.
Enhanced Reasoning and Tool Utilization
Claude 4 models can perform deep reasoning while using tools like web search through their expanded thinking capabilities. Compared to previous models, behaviors that exploit shortcuts or loopholes during task execution have decreased by 65%, providing more reliable results.
Claude Sonnet 4 achieves 72.7% on SWE-bench, showing significantly improved performance compared to the previous version, Sonnet 3.7. GitHub announced that they will introduce Claude Sonnet 4 as the new coding agent model for GitHub Copilot.
Strengthening Memory and Continuity
Claude Opus 4 can create and manage 'memory files' that store important information if developers provide local file access permissions. As demonstrated in cases like automatically writing a 'navigation guide' while playing Pokémon games, its ability to maintain context and utilize accumulated knowledge during long-term tasks has greatly improved.
Claude Code Officially Released
The Claude Code has been officially launched as a regular service after going through the research preview phase. You can check Claude's code editing suggestions directly in your IDE through beta extensions for VS Code and JetBrains, and it also supports background tasks via GitHub Actions.
Price and Usage Instructions
Claude Opus 4 is priced at $15/$75 per token (input/output), while Claude Sonnet 4 maintains the same price level as existing models at $3/$15. Both models are available through the Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI, with Sonnet 4 also available to free users.
As the advancement of AI technology accelerates, Claude 4 has improved to a level where it can provide practical assistance, especially to developers and researchers. Its ability to maintain consistent performance in complex coding tasks and long-term projects is expected to significantly expand the range of applications for AI assistants.
