Summary
Anthropic has released Claude Opus 4.7, an improved version of its AI model with better coding skills, clearer vision, and a new feature to check its own work. However, the company said this version still does not perform as well as its unreleased, more advanced Mythos model.
Key Facts
- Claude Opus 4.7 improves on the previous Opus 4.6 model in coding and visual tasks.
- Opus 4.7 performs better than ChatGPT 5.4 and Google Gemini 3.1 Pro on some tests.
- The Mythos model, more advanced than Opus 4.7, is only shared with select tech and cybersecurity companies and not publicly released.
- Users had complained that Opus 4.6 got worse recently, but Anthropic denied scaling down its abilities to save costs or shift resources.
- Opus 4.7 introduces a new effort level called "xhigh" for better control between problem-solving quality and speed.
- The company is testing "task budgets" to control how the AI handles longer tasks.
- Anthropic will use Opus 4.7 to trial safeguards to prevent misuse of the AI for cyberattacks.
- The lessons from Opus 4.7’s real-world use will guide the future release of Mythos-class models.
