ChatGPT (OpenAI) - capabilities
The default all-rounder and everyday workhorse: strongest at reasoning through complex documents, building reusable custom GPTs, and talking to AI by voice in the car. Our view from 500+ client engagements; capabilities evolve quickly.
Best at
- Reasoning models that break a problem into steps and course-correct
- Custom GPTs - reusable, shareable "your own version of ChatGPT" assistants
- Advanced voice - hands-free, two-way AI on the move
- Handling very large documents via retrieval (RAG), where full-context models choke
- Being the broad daily default for research, writing and Q&A
Capability snapshot
| Capability | Verdict | What that means |
|---|---|---|
| Reasoning | ✅ Leads | His canonical "thinking model" - takes time, breaks the problem down and self-corrects on complex documents. |
| Custom assistants / GPTs | ✅ Leads | Custom GPTs store and reuse prompts; easy to build, shareable across a company, with knowledge files. |
| Voice | ✅ Leads | Advanced voice is a favourite - two-way speech in the car to draft emails and do research. |
| Web search | ✅ Strong | Browses live, even inside voice mode, to pull up information on the fly. |
| Long documents / context | ✅ Strong | Retrieval (RAG) lets it tackle huge files that full-context models can't - but scanning can miss detail. |
| Deep research | ✅ Strong | Agentic research across many sources into a cited report. |
| Image generation | 🟡 Capable | Generates images in-line; capable, not its headline strength. |
| Office docs (Excel / PowerPoint / Word) | 🟡 Capable | Reads and works with Office files well; for building polished deliverables we lean on Claude. |
| Code & data analysis | 🟡 Capable | Reads and reasons over PDFs, spreadsheets and CSVs to answer questions. |
| Memory | 🟡 Capable | Remembers context across a conversation - sometimes a little over-eagerly. |
In Wouter's words
ChatGPT also has reasoning models - they call them O models, O3. You will see that the answer of a reasoning model is way more thought through than the answer of a normal model.
Custom GPTs are actually your own version of ChatGPT. I have an email maestro… in the background it has a very elaborate instruction of how I write emails. So this is like my blueprint of my emails.
Watch-outs
- Retrieval on big documents means it doesn't read every word - risky for exhaustive "list ALL the items" questions.
- For producing finished Office deliverables (Word/Excel/PowerPoint), we tend to favour Claude.
- Outputs are first drafts you must check - it still makes mistakes.
- The free tier may use your data for training; use the paid/Team workspace for real company data.
Our take
ChatGPT is the dependable all-rounder and our most common reference point: the model to reach for first for reasoning through complex documents, for building reusable custom GPTs, and for talking to AI by voice. Its retrieval approach lets it handle enormous documents that full-context rivals can't, with the honest caveat that scanning can miss detail. When the job is producing a polished Office file, we'll often hand off to Claude.
Voice is my favourite capability - when I drive I speak to an AI for an hour, and it does all my research.
- Wouter van Haaften, WAIMAKERS
Compared head-to-head