HISTORICAL RELEASE COMPARISON
GPT-4 vs GPT-3.5: what changed?
GPT-4 added stronger handling of complex instructions and demonstrated image understanding. Original ChatGPT made conversation the interface for a GPT-3.5 model.
These are the releases named below, not current product recommendations.
| What to compare | ChatGPT (GPT-3.5) | GPT-4 |
|---|---|---|
| Snapshot | November 2022 ChatGPT | March 2023 GPT-4 |
| Main change | Conversational follow ups | More complex instructions and visual inputs |
| Image access | Text focused launch | Research preview at announcement |
What this means in practice
For a simple rewrite, a stronger benchmark result may not change the experience much. For a task with several constraints, compare whether the model follows all of them. The task matters more than a universal “times smarter” label.
Is GPT-4 ten times smarter?
The sources do not establish a universal intelligence multiplier. A difference on one exam cannot be converted into the same improvement on every task.
Read the individual model stories
Sources
Developer reports describe their own results. This site has not independently reproduced them.
OpenAI: Introducing ChatGPT OpenAI: GPT-4 launch and availability OpenAI: GPT-4 Technical Report