GPT 5 Shipped

BLUF Nothing earth shattering. AI is not taking over the world. It is a lot cheaper and a bit safer.

What did not change

What actually changed

Safe Completions - OpenAI trained GPT 5 to focus on the safety of its output rather than guessing your intent. Instead of a hard refusal, you will often get a constrained answer with an explanation and safer alternatives. This is better for dual use questions like security topics where context matters.

Price and Scale - In the API, GPT 5 lists a 400k token context and meaningfully lower prices than past flagships. That drops the cost bar for real work, including long docs and multi step coding tasks. Performance that was once behind "pro" paywalls is now widely available. (This is the best part of the update you can click off now.)

Access and voice - GPT 5 is now the default in ChatGPT with plan based limits, and the improved voice experience is expanding to more users, including Free. Again, premium features are now becoming standardized.

Benchmarks - OpenAI claims better coding and fewer hallucinations than o3. Treat this as a prompt to run your own evals on your data and tools. My results did not differ greatly from the prior models' behavior and I would maintain the same level of skepticism.

Reasoning is slightly better but not magical. You still need human review for sensitive tasks. In my testing, GPT5 reasoning surpasses 4o but not o3. It seems we will have to wait for GPT6 to deliver the advanced reasoning speculators have hoped for.