sui@birdabo
Claude Opus and Sonnet finally got 1M context window.
> can load entire codebases, 1000+ pages, 600 images/pdfs pages - all at once with minimal context loss.
> at this scale, GPT-5 drops to 36% accuracy, Gemini to 25%, but Claude stays at 78%
we’re maybe 40% to AGI now.
Opens with a number
