Every documented ChatGPT limitation in 2026 with version, source, and date: hallucination splits, usable context, memory recall, coding regressions.

GPQA Diamond went from 38.8% to ~94% in under three years. What the benchmark measures, where scores stand in July 2026, and why the last 6% may be flawed questions.

How ChatGPT, Gemini, and Perplexity assemble brand recommendations: retrieval pipelines, citation data from 230K prompts, and what the evidence says moves visibility.

A running, sourced ledger of documented failure modes in frontier AI tools: hallucination rates, context rot, agent reliability, and insecure code.