GPT-5.2 and GPT-5.2-Codex are now 40% faster

OpenAI dropped a mic on X: GPT‑5.2 and its coding cousin are now “40% faster” with the same model and weights — basically, quicker replies at the same price. Cue the fireworks. The dev-hype squad showed up first. One coder bragged that recent updates — sub‑agents, a new multi‑agent app, plus this speed bump — have tripled productivity. The vibe: “Ship day, every day.”

Then the skeptics stormed in. A loud chorus swore the model’s been quietly nerfed since launch, with one user claiming the early shine was just to impress journalists and benchmarks. Another pointed to “tons of posts on Reddit” about quality dropping. The spiciest subplot? A running theory that staff accounts get the fast lane, while regular users languish in traffic — complete with a commenter linking to a think piece and muttering, “normal people have it slower than staff.”

Even the money angle got messy. One commenter wondered why the price didn’t budge if the underlying hardware (like Cerebras chips) is pricey — a subtle “who’s really paying for this?” jab. Memes flew: “Speed for who?” “Sonic the Hedgehog, but only with a badge,” and “Fast Mode vs. Fast One.” TL;DR: the timeline split between speed-celebration and trust-suspicion, with OpenAI’s “same weights, lower latency” line becoming the day’s Rorschach test.