New peak/off-peak billing structure took effect August 16, pushing output tokens from $0.87 to $3.96/million during peak hours; off-peak rates also rose, though V4-Pro remains cheaper than Western equivalents.
Why it matters: Teams using DeepSeek V4 as a cost-efficient backbone need a budget review — peak-hour output costs jumped ~4.5x, and the new peak/off-peak spread means usage timing now matters for cost control.
Three Imagen 4 model IDs (standard, ultra, fast) are retired as of August 17; the replacement Gemini 3.x Image API uses generatecontent(), not generateimages(), so any unmigrated app is broken right now with no fallback.
Why it matters: Any stack calling Google's image generation API is broken as of today — the shutdown is a hard cutoff, not a soft deprecation, and the migration requires reworking the API call pattern, not just swapping a model ID.
ChatGPT's extended-reasoning mode is now available to free users, while Plus/Pro subscribers get a per-response effort slider to trade speed for depth on demand; an updated GPT-5.6 Sol and a Linux desktop public preview also shipped August 15.
Why it matters: Reasoning capability is no longer a paid-tier differentiator — free ChatGPT users now get the same quality floor as paying subscribers, narrowing the gap and raising the baseline for AI-assisted content and analysis work.