Deep DIVE: AI progress continues, as IQ scores rise linearly (www.maximumtruth.org)

TrackingAI.org’s month-by-month benchmarks show AI “IQ” rising steadily rather than stalling: the top model moved from the mid‑80s to about 130 in roughly 1.5 years, with an average gain of ~2.5 IQ points per month from May 2024 to Oct 2025. The site uses four measures — including an “offline” test explicitly kept out of training data and the public Norway Mensa problems — and finds the leading systems already near ceiling on Mensa (ChatGPT Pro has twice scored 34/35 → ~148) while scoring ~130 on the offline benchmark. Vision-capable models have closed a gap too: about a year ago they began solving image puzzles directly, with “vision” IQs around 105–112 (ChatGPT Pro averages ~105), and ChatGPT 5 appears to have contributed a notable vision improvement. Grok and Anthropic’s Claude 4 Opus recently competed for the top spot, with Grok raising the top score from ~120→130. Why it matters: the data argue AI reasoning is improving linearly and predictably, not plateauing — implying more steady gains that should continue for years. Technical caveats include potential overlap of public Mensa items with training data and remaining disparities between reasoning and visual perception. The author projects leading AIs will master visual IQ tests by late 2027, suggesting a manageable window for adaptation as models increasingly supplement human tasks before any broader agency concerns arise.