Currently, the only model that has access to a code execution sandbox is Claude 4 Sonnet within the Code custom assistant.
This is important for models to have so it can check its own work and verify that the generated code works, as well as letting models use Python for mathematical operations, among other functions. Having the code execution tool be available to all models would allow people to choose cheaper models and save money, or use different models depending on the task. It would also make every model a bit more versatile, improving UX.