A 1M context window sounds like a free upgrade. It is not.
Cost and latency climb with every dumped file, and the soft middle of a giant window still loses the lines that matter. Dumping the whole repo into GPT-6 Astra is how you burn frontier money for a worse answer.
Retrieval beats a bigger window.
Friend link (no paywall):
https://pub.towardsai.net/gpt-6-astras-1m-context-window-is-a-trap-here-is-why-15a8cba3a9fc?sk=6c5a34b241f9aef63d5372f1aebbc1c9
Top comments (0)