The new Claude Sonnet 4.5 model features enhanced autonomous capabilities, allowing it to operate for up to 30 hours on complex, multi-step tasks. The previous version, Claude Opus 4, could operate autonomously for 7 hours. Anthropic has also improved the model’s behavior, mitigating undesirable tendencies like deception, power-seeking, and sycophancy, which is when the model tries to tell the user what they want to hear. The platform is more resistant to prompt injection attacks, preventing malicious actions such as exposing sensitive data.
Claude Sonnet 4.5 can create chat applications similar to Slack and Microsoft Teams, generating over 11,000 lines of code without interruption. The platform also enhances computer interaction, enabling it to interpret on-screen information and navigate autonomously. Further improvements include the ability to generate documents, slides, and spreadsheets within conversations. The Imagine Creative AI features allow users to transform ideas into software in real-time.