Anthropic introduced Claude Sonnet 4.5 — it can work autonomously for 30 consecutive hours
And the Claude Code programming service has added a "checkpoints" feature, which allows you to save task progress and return to it.
- Anthropic announced the updates in a blog post. The company calls Claude Sonnet 4.5 the most powerful model for programming and creating AI agents. It also demonstrates significant progress in reasoning and mathematics.
- Claude Sonnet 4.5 is capable of working autonomously for 30 consecutive hours. According to Anthropic, during tests, the model wrote about 11,000 lines of code to create a Slack-like application and only stopped after completing the task.
- The new model is especially effective in cybersecurity, finance, and research, the company claims. Beta testers, including Cursor, GitHub, Canva, Netflix, and others, note that Sonnet 4.5 is better at code editing, vulnerability scanning, product research, and other tasks with long context.
- Agents based on Claude Sonnet 4.5 are more than three times better at browser navigation and using a computer on behalf of a user, Diane Penn, head of product management at Anthropic, told The Verge.
- The model can operate at an executive level—for example, by checking the calendars of multiple employees and scheduling meetings, as well as gathering analytics and updating task information following meetings with subordinates, added Scott White, head of product for Claude.ai.
- In addition, Anthropic has updated Claude Code. It now uses Sonnet 4.5 by default. The service has added a checkpoints feature, allowing users to save task progress and revert to it, undoing changes in the code.
- And it handles very good with svg creation, check this simple prompt and result:
