Developer's first Astra hands-on: better browser control, faster quota drain
A developer says Astra is much better at browser and computer use and faster than 5.6 sol, but only the low reasoning mode is practical and quota drains 2-3x faster.
First thoughts on GPT 6 (aka Astra): - Model is so much better at browser/computer use - It feels faster than 5.6 sol even without fast mode - Only usable reasoning mode is "low", otherwise limits go down way too quick - Way better at UI, still not Claude level but like 85% of the way there Also I highly recommend you still delegate tasks to 5.6 sol on low/medium, Astra is great but 5.6 sol can do boring stuff like looping with your AI code review agent and fixing issues. It would be really cool to see OpenAI optimize/raise limits on current plans, I burnt through my limits like 2-3x faster because I used it on medium thinking.