back

The new model posted impressive benchmark scores for coding, beating out GPT 5.1 and Gemini 3.