The general-purpose model has won the coding war. OpenAI's decision to merge Codex into GPT-5.5 — eliminating the standalone coding agent line entirely — is not a product reorganization. It is an admission that the specialized coding agent era is over.
Romain Huet confirmed the unification on April 25th: since GPT-5.4, Codex and the main model operate as a single system. GPT-5.5 extends this further with "strong gains in agentic coding, computer use, and any task on a computer." There will be no GPT-5.5-Codex. The separate coding line that once promised specialized benchmark dominance is gone.
This matters because OpenAI spent years cultivating Codex as the answer to code-specific benchmarks. The company's internal narrative positioned specialized agents as the path to coding supremacy. That narrative is now dead. The general model absorbed Codex's capabilities — and OpenAI is betting that this unification outperforms the fragmented approach.
OpenAI's own prompting guide for GPT-5.5 makes this explicit. Developers should "treat it as a new model family to tune for, not a drop-in replacement." Start from scratch. Do not carry over existing prompts. The company is essentially telling practitioners: the old playbook is obsolete. GPT-5.5 requires fresh baselines.
This advice reveals something uncomfortable for teams still building specialized coding pipelines. If the flagship general model now handles agentic coding better than dedicated solutions, the economic case for specialized agents weakens. OpenAI has effectively conceded that capability gains now come from general improvements, not domain-specific tuning.
The practical implication is stark: developers who built workflows around standalone Codex should expect migration friction. OpenAI's upgrade command — `$openai-docs migrate this project to gpt-5.5` — exists precisely because the model behavior has shifted. Existing prompts optimized for earlier models will not translate cleanly. This is not a minor versioning quirk. It is a fundamental architecture change disguised as a product update.
For the broader ecosystem, the signal is clear. If OpenAI — the company that pioneered specialized coding models — abandons that strategy, competitors will feel pressure. The benchmark war is over. General models won by becoming good enough at everything that dedicated agents became redundant.