Hacker News
A Cost Analysis Puts the Halving Time for Index-60 Intelligence at About 34 Days, With Grok 4.6 Holding the Record at $0.84 Per Task
A CatalystNeuro post that drew 125 points argues LLM cost curves follow a repeating premium-debut-then-commoditization shape per capability tier, with records halving every 4 to 10 weeks. Cited figures as of 2026-08-19: intelligence index 40+ fell 56x between February and August 2026, from Claude Sonnet 4.6 at $1.22 per task to GPT-5.6 Luna at $0.022; index 50+ fell 35x in five months; index 60+ fell 3.8x since June, halving roughly every 34 days, with Grok 4.6 at $0.84 per task against Claude Opus 5 at $2.34 on max effort. The builder implication the author draws is that read-everything workloads flip from uneconomic to routine, and total spend rises anyway.
↳ Follow the thread