LLMs are deterministic if you force a seed or disable sampling. They however do not guarantee that small input changes will cause small output changes.
Not with todays GPU's, you would need to run it locally with special GPU settings or run it on your CPU to ensure it is deterministic.