Claude vs. Gemini: Testing on 1M Tokens of Context