ChatGPT's April overhaul: File Library, GPT-4o retirement, and a cleaner model lineup

What You'll Learn

• Which model replaces GPT-4o for everyday tasks (and which handles complex reasoning)
• How to migrate custom GPTs and API integrations off deprecated models
• How to set up ChatGPT File Library to eliminate re-uploading forever
• Why OpenAI's next model (Spud) is worth monitoring this month
• How Codex-only seats in ChatGPT Business change the licensing math for dev teams

OpenAI shipped three meaningful changes to ChatGPT in the first week of April. Not features (s)tructural changes to how the product is organized and how it stores your work. GPT-4o is gone. The file you uploaded last Tuesday is now saved. And the model selection menu finally matches the way people actually think about tasks.

If your team runs custom GPTs, API integrations, or any workflow built around ChatGPT, some of this requires action now. If you're a daily ChatGPT user who just uploads documents and gets answers, most of this is good news.

The new model lineup (what each one is for)

GPT-5.3 Instant replaces GPT-4o for everyday speed tasks: writing drafts, summarizing documents, answering questions, editing copy. It's faster than Thinking and significantly cheaper at scale. Use it for anything that doesn't require deep reasoning or extended analysis.

GPT-5.4 Thinking handles complex reasoning: financial modeling, strategy documents, research synthesis, code debugging, multi-step analysis. It takes longer than Instant and costs more per query, but the quality difference on hard problems is measurable.

GPT-5.4 Pro is the highest-capability option: use it for research-grade outputs, lengthy document analysis, technical writing that requires extended precision. Pro is not for everyday tasks (t)he cost-per-query is significantly higher than Thinking.

How to migrate custom GPTs and API integrations

Any custom GPT that was created with GPT-4o as the base model needs to be updated. Open the custom GPT editor, find the model selection setting, and switch to GPT-5.3 Instant (for utility and speed tasks) or GPT-5.4 Thinking (for reasoning-heavy GPTs).

For API integrations: search your codebase for "gpt-4o" and update to the appropriate successor. Most gpt-4o use cases map to gpt-5.3-instant. Test before deploying (o)utput format and tone can shift across generations. Budget a testing session to verify outputs before going live with the updated integration.

ChatGPT File Library (s)etup and use

Every file you upload to ChatGPT (Plus, Pro, or Business) now saves automatically to the Library in the left sidebar. To use it:

• Upload any file during a chat (i)t saves to Library immediately.
• In any new chat, click the attachment icon and select "Library" instead of uploading a new file.
• Search by description: type "Q3 budget" or "client proposal" and Library surfaces matching files by name and content.
• Filter by type: Documents, Spreadsheets, PDFs, Presentations, Images.
• Files stay in Library until you manually delete them.

What doesn't save: files uploaded in Temporary Chat mode. Turn off Temporary Chat if you want Library to capture everything. Currently web-only at chatgpt.com. EEA, Switzerland, and UK users are waiting on a separate rollout.

The GPT-5.5 Spud situation

OpenAI's next major model completed pretraining around March 24, 2026. It's internally codenamed 'Spud' and expected to release as GPT-5.5 or GPT-6 between April 14 and May 5. Polymarket gives this a 78% probability.

My take: don't hold workflow decisions waiting for Spud. The current 5.4 generation handles everything most executive teams need today. If Spud ships with a meaningful capability jump, it'll show up in the benchmarks fast. Adapt then.

Codex-only seats for ChatGPT Business

OpenAI now offers Codex-only pay-as-you-go seats for ChatGPT Business. This means developer team members get access to Codex coding capabilities without requiring a full Business seat. If your team has a mix of knowledge workers and developers, right-size licensing now rather than paying full price for engineers who only need code assistance.

Action Steps Summary

• Audit custom GPTs. List all custom GPTs in your workspace. Note which ones use GPT-4o or earlier models. Check their creation date to estimate migration effort.
• Update API integrations. Search your codebase for "gpt-4o" references. Create a test environment and swap models one integration at a time, running output tests before deploying to production.
• Populate File Library. Upload 3 reference documents (your most-used PDFs, templates, or templates) this week and test retrieval. Confirm they appear in your Library across new chats.
• Set the right default model per task. For routine work, default to GPT-5.3 Instant. For analysis and synthesis, use GPT-5.4 Thinking. Document this in your team's ChatGPT usage guide.
• Evaluate Codex-only seats. If your team includes developers, request a licensing audit. Calculate cost savings from switching engineering team members to Codex-only vs. full Business seats.