OpenAI raises ChatGPT custom-instructions limit to 5,000 characters for paid users

OpenAI has more than tripled the custom-instructions ceiling in ChatGPT, from 1,500 to 5,000 characters, for users on Pro, Enterprise, Business, and Education plans. Custom instructions let users persistently encode preferences — tone, formatting, domain context, do's and don'ts — that apply automatically across all conversations, and the expanded budget allows far more nuanced, stable behavior without repeating context each session.

The change is retroactive, meaning existing chats benefit immediately, and OpenAI frames it as coinciding with a model that handles these longer instruction blocks more reliably. It's a small-but-meaningful power-user feature: enterprises building repeatable workflows and individuals with detailed style guides gain headroom that previously forced awkward trimming.

The move fits OpenAI's busy week — full GA of the GPT-5.6 Sol/Terra/Luna tier, ChatGPT Work agentic launch, a $230 Codex hardware companion, and reports of a first physical device — as the company races to deepen its paid tiers amid an intensifying AI price war (OpenAI, Meta and Musk all slashing model costs). Bigger custom-instruction budgets are a retention lever that raises switching costs.

Ethan Mollick's viral caution is relevant here: he warned that AI-written 'anti-slop' instruction files can hyperstition outputs into 'mega-slop' unless the author has genuine taste — a reminder that more instruction space amplifies both good and bad prompting. What to watch: whether OpenAI extends the limit to free users, and whether longer instructions measurably improve consistency or simply give users more rope to over-specify.