OpenAI's GPT-5.4 Can Now Hold Your Entire Manuscript in a Single Conversation

You can now paste an entire novel into ChatGPT and it won’t forget what happened in chapter three. That alone makes GPT-5.4 worth talking about.

OpenAI released GPT-5.4 on March 5 with a 1 million token context window. That’s roughly 750,000 words, or about three full-length fantasy novels, in a single conversation. If you’ve ever wished you could dump your whole manuscript into ChatGPT and ask it to find continuity errors or flag every scene where a character’s eye color mysteriously changes… you can do that now. No more breaking your text into chunks and losing context between sessions.

What Changed from GPT-5.3

We covered GPT-5.3 Instant last week, which focused on writing quality improvements and fewer frustrating refusals. GPT-5.4 is a different beast aimed at a different problem.

The key numbers: 33% fewer factual errors compared to GPT-5.2 on individual claims, and 18% fewer errors overall across full responses. For authors who use ChatGPT to research historical details or verify publishing industry info, that’s a real improvement in reliability. Fewer hallucinations means less time double-checking the AI’s homework.

GPT-5.4 also comes in three versions. Standard, Pro (optimized for complex tasks), and Thinking (which shows its reasoning process before answering). The Thinking version lets you see how the model plans its response, which could be useful for authors working through plot problems or structural revisions where you want to steer the AI’s approach.

What Authors Should Know

The context window is the big deal. Previous models capped out around 128K-200K tokens, which meant long novels had to be split up. Splitting means the AI loses track of details from earlier sections. A 1 million token window eliminates that problem for virtually any single manuscript.

Pricing matters. The standard model runs $2.50 per million input tokens and $15 per million output tokens via the API. Requests over 272K tokens are charged at double rate. If you’re using ChatGPT directly (not the API), your access depends on your subscription tier. If you use BYOK tools like NovelCrafter, you can access GPT-5.4 through the API now, though feeding a full novel through it won’t be cheap.

This is more analyst than co-writer. GPT-5.4’s improvements lean toward accuracy and handling large inputs. If you’re looking for better prose quality in AI-assisted drafting, GPT-5.3 Instant (which is already the default in ChatGPT) is still the more relevant update. GPT-5.4 is the model you bring in when you need a sharp reader who can hold your whole book in their head at once.

Worth Trying

If you have a finished manuscript, try uploading it and asking GPT-5.4 to identify continuity issues or flag scenes where tension drops. These are tasks that benefit enormously from the model seeing the full text rather than isolated excerpts. No AI is a substitute for a human developmental editor (not yet, anyway), but as a first-pass analysis tool? This is the most capable option we’ve got.