- Add Gemini 4 Carbon (High) with 1M context length and 1M max completion tokens.
- Include thinking levels (low, medium, high) with dynamic thinking support.
- Support text, image, audio, and video input modalities with text output.
- Enable web_search native capability for the new model.