Explore other topics:hassabis deepseekdeepseek model sizesdeepseek v2 architecturesatya nadella deepseekdeepseek r1 vs gemini 2.0 flash thinking