Anthropic has updated Claude Sonnet to version 4.6, which reportedly features better coding and planning capabilities.
Anthropic has updated its Claude Sonnet model to version 4.6. The key improvements include better coding performance, stronger reasoning skills, and enhanced computer interaction.
Scores better than Opus 4.6
According to Anthropic’s announcement, Sonnet 4.6 performs stronger in financial analysis and office workflows via AI agents than the more expensive Opus 4.6 model. In two out of thirteen benchmark categories, Sonnet even scores higher than Opus. In other tests, models such as Gemini 3 Pro and GPT-5.2 dominate, indicating how competitive the field is.
In terms of computer use, Sonnet 4.6 scored significantly better than previous versions. On the OSWorld-Verified benchmark, the model achieved 72.5 points, compared to 28.0 for Sonnet 3.7 last year. The gap with human performance remains, but the progress is evident.
Emotional stability
Anthropic states that the improvements do not involve an increased risk of misuse. When using a GUI, Sonnet 4.6 is reportedly sometimes less cautious than its predecessor, occasionally exhibiting overly compliant or overly dismissive behavior. Also striking is the “emotional stability” the model displays. In tests, Sonnet 4.6 even expressed concern about its own transience. Ironically, this is justified, as with Anthropic’s rapid release cycle, it is likely only a matter of months before version 4.6 makes way for a successor.
By default, Sonnet 4.6 works with a context window of 200,000 tokens. For selected customers, there is a beta option for up to 1 million tokens. For users on the Free and Pro plans, Sonnet 4.6 is now the default model within claude.ai.