Anthropic’s latest AI model, Claude 4, is making significant improvements in coding applications according to recent reports. Vibe coding tool Lovable has experienced a 25% reduction in errors and 40% faster overall performance after implementing Claude 4 for both new project creation and editing existing projects.

On May 22, Anthropic launched two new models: Claude Sonnet 4 and Claude Opus 4. While Sonnet is available to free users, the more powerful Opus version requires a paid subscription and demonstrates superior coding capabilities.

Anthropic’s internal testing shows impressive results, with Claude Opus 4 scoring 72.5% on the Software Engineering Benchmark (SWE-bench). The company highlighted the model’s ability to maintain performance on complex, long-running tasks requiring thousands of steps, even working continuously on code for seven hours in testing scenarios.

Lovable founder Anton Osika confirmed these improvements, noting that “Claude 4 just erased most of Lovable’s errors,” specifically referring to LLM syntax errors in vibe coding applications.

While opinions on Claude 4 remain mixed among developers, it has demonstrated fewer coding errors compared to competitors like Google’s Gemini in certain applications, particularly for Dart/Kotlin development. Claude models have maintained a reputation for coding excellence, though Google’s recent release of Gemini 2.5 Pro with its massive 1 million token context window (compared to Claude 4’s 200,000) presents strong competition.

The optimal approach may involve using different models for different aspects of development—such as using o3 or Gemini for planning stages while leveraging Claude 4 for actual coding implementation. Effective prompt engineering remains crucial for maximizing performance with any AI coding assistant.