GPT‑5.4 rumor roundup

Rumors say some ChatGPT Pro users are seeing GPT‑5.4 Pro run much faster and hold up to longer, harder prompts than before. (aihola.com) Users posting on X and in Pro communities reported response times for “hard” queries dropping from roughly an hour to about 15–20 minutes, and some described noticeably sharper code and SVG outputs. (aihola.com) Unofficial spec lists and chatter claim an approximate ~3x speed gain for Pro, a Mini variant with a 400,000‑token context window, and endurance for 15–20 minute single responses on heavy workloads. (aihola.com) OpenAI’s March 5, 2026 announcement says GPT‑5.4 supports long‑horizon work and integrates Codex-style coding/agent features, and the company reports fewer errors versus GPT‑5.2 in its published benchmarks. (openai.com) Those official materials list a 1,000,000‑token context for the main GPT‑5.4 model and describe “Thinking,” “Pro,” and “Instant” variants for different workloads. (openai.com) The timing of the chatter centered April 21–22, 2026, as Pro subscribers compared examples and speed tests; reporting notes OpenAI had not publicly confirmed the specific 3x or 15–20 minute metrics. (aihola.com) OpenAI’s community forum and docs show a /fast mode that can deliver up to about 1.5x faster inference in some cases, which is smaller than the 3x figure users are circulating. (community.openai.com) Separately, GitHub repos and forum posts have circulated alleged system prompts and internal files suggesting model‑level changes and Codex/ChatGPT alignment, but those artifacts lack independent verification. (github.com) For context, OpenAI first introduced GPT‑5 on August 7, 2025, and followed with the GPT‑5.4 family in March 2026 as the company pushed for longer, tool‑heavy workflows. (openai.com) OpenAI has not issued a model‑card update confirming the specific speed or endurance numbers, and watchers say an official update or testable API change would be the clearest way to verify the rumors. (aihola.com)