GPT-5.4 hits Codex with 1M context

This looks like a bigger deal for coding workflows than for chatbot novelty. OpenAI is pushing one frontier model closer to a true production engineer assistant instead of a clever autocomplete engine.

• –Experimental 1M context in Codex could cut down on the retrieval, chunking, and reprompt gymnastics that large-repo work usually needs
• –128K output gives the model more room to emit real artifacts like multi-file patches, tests, migrations, and long structured plans
• –Shipping GPT-5.4 across API, ChatGPT, and Codex reduces the usual split between “prototype in chat” and “actually use it in tooling”
• –Early hands-on coverage frames it as noticeably better for production-style coding tasks like full website generation, even if reliability quirks have not disappeared
• –The real question now is whether the long context stays useful under real load, not just in launch demos