1/ 🚨 GPT-5 is officially live and debuts at #1 on LMarena. Four variants, GPT-5, mini, nano, and chat launch with unified reasoning and agentic upgrades. Even free users get the frontier model. #GPT5 #AI

3/ Claude Code is still our gold standard at Array AI Labs but GPT-5 claims “beyond vibe coding”. But the real test: can it ship code, not just snippets? We’ve put it to some early tests and our thoughts are •⁠ ⁠Points out your mistakes, doesn’t just follow prompts. •⁠ ⁠Switches and questions direction/context mid-task •⁠ ⁠Handles larger (repo-scale) jobs •⁠ ⁠Finally feels like a senior dev, not a mid-level assistant •⁠ ⁠Major leap in frontend/design suggestions over Codex and older models

4/ Cost matters: GPT-5 costs less than Claude Opus 4.1 for similar performance in many use cases. But as X is showing, some users say “Claude still feels sharper for complex refactor tasks.” GPT-5-mini is a hit for builders i.e. cheaper, but nearly as capable as 2.5 Flash

5/ The takeaway for founders & teams: Massive context = you can dump product docs, interviews, even (some) codebase in one go True agentic capability and error correction for daily use, not just demos. But remain skeptical and try both, benchmark on your stack, and show your receipts. Comparison threads coming from our team.