Lead AI Friday: GPT-5 just dropped!

Here's your executive summary

• Rolling out today for Free/Plus/Pro users, Enterprise gets it next week

• All previous models will be discontinued

• GPT-5 picks by itself whether to answer fast (like 4o) or think deeply (like o3)

Pricing changes (API)

• GPT-5: $1.25 input / $10 output per 1M tokens

• That's 25% more expensive than GPT-4o for input, same for output

For programming

• Available in Cursor now with free tokens during launch - a good time for A/B testing

• 74.9% success rate on SWE-bench (vs Claude 4.1 Opus's 72.5%, Gemini 2.5 Pro's 63.2%)

• Runs multiple tools at the same time → faster results

• Better at cross-file reasoning and long coding tasks

What caught my attention

• 26% fewer hallucinations compared to GPT-4o

• Gmail and Calendar access (Pro tier first) - can't wait to give it a spin!

• Much better at following instructions (99% on tests vs ~85% before)

Available in Cursor and OpenAI Playground now. ChatGPT access is rolling out gradually.

I'll be testing it heavily over the next days and will share what I've found later.

What's your experience with GPT-5 so far? Have you noticed the improvements?