Frontier Model Releases Are Now Shipping on a Software-Patch Cadence, Not a Product-Launch One

August's model releases arrived close enough together that comparing them head-to-head is becoming its own genre of coverage rather than a rare event. GPT-5.6 Luna, a mid-cycle update rather than a full version bump, DeepSeek's V4-Flash-0731 update, Meta's Muse Spark 1.1, and Thinking Machines' Inkling all shipped within roughly the same window — and independent tracking of major-lab releases now puts the cadence at close to four times where it stood in 2023. The practical effect for anyone building on these models is that version-pinning has become a real operational discipline rather than a formality. A model that's the best available option in July can be a generation behind by September, and evaluation suites that took a week to run against one release now need to run continuously just to know which model is actually ahead this week. For smaller teams, the upside of the pace is real: Luna's release specifically emphasized aggressive price cuts alongside capability gains, following DeepSeek's earlier undercutting of Western pricing — meaning the cost of staying current is falling even as the frequency of needing to re-evaluate rises.