my thoughts on gpt-5

been using gpt-5 (max) inside cursor for the past two days and here’s my very early reaction:

• it follows instructions with surgical precision
• tool calls are clean and to the point; i’ve seen it do 11 file edits in one go
• for long tasks, it actually stays on track without drifting
• it doesn’t ramble or explain unless asked, just quietly gets stuff done
• writes code that works, no babysitting needed

the real difference shows up when you run todo-style breakdowns or assign multi-step work. it just finishes the task. cleanly.

Example: one of our engineers saw tens of thousands of calls to a legacy GET /me/... endpoint slowing down page load and asked gpt-5 to remove it. it deleted ~3,000 lines across the codebase and rewired things cleanly. no notes, just results.

on the flip side, it still shares the same issues gpt-4o always had when reasoning deeply in code. gpt-5 thinking feels closer to o4 than to a leap forward, which is both good and limiting.

used exclusively for code, 70% of the time inside cursor and the remaining via the cursor cli (clisor), powered by gpt-5 max (272k context window and medium reasoning effort), nothing fancy. but for my day-to-day dev work, this is already starting to feel like a serious upgrade.

this is just a first look. real judgement will come after shipping a full project with it or solving a complex bug.

• biggest diff with sonnet is i find it follows direction better
• it doesn't make quite as many mistakes; even if sonnet corrects itself, it causes a lot of context pollution

my experience so far: can't tell the difference between sonnet and gpt-5. it’s basically a slightly better sonnet for cheaper.