Pattern: AI Coding Price War Enters a New Phase — 6x Cost Gap Between Open and Closed

VentureBeat·high signal

DeepSeek V4-Pro at $3.48/M output tokens vs GPT-5.5 at $30/M and Opus 4.7 at $25/M creates a 6-8x price gap while approaching parity on SWE-bench (80.6% vs ~80.8%). Simultaneously, GPT-5.5 doubles prices from GPT-5.4 levels. The market is bifurcating: frontier labs raise prices banking on capability moats, while open-weight models close the gap at commodity costs. For builders, this means model routing isn't optional anymore — use expensive models for hard reasoning, cheap ones for everything else.