Krux
March 30, 2026
ChatGPT Cuts Factual Errors by 33% in Latest Update
Published: March 30, 2026 at 1:47 AM
Updated: March 30, 2026 at 1:47 AM
100-word summary
ChatGPT's March update delivers 33% fewer factual errors and expanded memory that persists across sessions. Translation: the AI can remember last week's conversation and hallucinate less when you ask it to draft a report or analyze data. The upgrade, GPT-5.4 Thinking, is live now for Plus, Team, and Pro subscribers. Long research projects can finally maintain coherence without re-explaining context every time, and knowledge workers spend less time fact-checking outputs. OpenAI is retiring older GPT-5 variants to push everyone onto the new model. The timing isn't random: Anthropic just launched an institute studying AI's impact on jobs, signaling the industry knows unreliable tools won't survive workplace scrutiny.
What happened
ChatGPT's March update delivers 33% fewer factual errors and expanded memory that persists across sessions. Translation: the AI can remember last week's conversation and hallucinate less when you ask it to draft a report or analyze data. The upgrade, GPT-5.4 Thinking, is live now for Plus, Team, and Pro subscribers. Long research projects can finally maintain coherence without re-explaining context every time, and knowledge workers spend less time fact-checking outputs.
Why it matters
OpenAI is retiring older GPT-5 variants to push everyone onto the new model. The timing isn't random: Anthropic just launched an institute studying AI's impact on jobs, signaling the industry knows unreliable tools won't survive workplace scrutiny.