Codex upgrades repos to GPT-5.5

Hot take: this is less a product reset than OpenAI turning Codex into the delivery layer for model upgrades. If GPT-5.5 is really more efficient and more reliable on real repos, that matters more than another generic “smarter model” claim.

• –OpenAI frames GPT-5.5 as stronger on complex terminal workflows, GitHub issue resolution, refactors, and code review, which is the stuff teams actually pay coding agents for.
• –The efficiency angle matters: fewer tokens and fewer retries can turn a nice demo into something viable for long-running repo work.
• –This is strategically important because Codex becomes the place where model improvements land directly in developer workflows, rather than forcing teams to re-evaluate their stack.
• –The competitive pressure is obvious: Cursor, Claude Code, and GitHub Copilot all need to justify why their coding loops are better than an OpenAI-native one.