OpenAI kills its dedicated coding model Codex again, folding it into GPT-5.5

Editor's take

OpenAI has retired its specialized Codex model, integrating its coding functionalities into the upcoming GPT-5.5. This move suggests a shift towards more generalized, highly capable LLMs rather than siloed, task-specific AI. The aim is to imbue the core model with advanced agentic coding abilities and improved efficiency, potentially impacting developers who relied on Codex for specific coding tasks and now need to adapt to a more integrated AI.

The significance lies in OpenAI's continued consolidation of AI capabilities into its flagship models. This strategy, seen previously with the deprecation of earlier Codex versions, indicates a belief that a single, powerful LLM can outperform specialized models. The success hinges on GPT-5.5 delivering demonstrably superior coding performance and efficiency compared to its predecessors, justifying the discontinuation of dedicated tools.

Future developments to monitor include the actual coding benchmarks and agentic capabilities of GPT-5.5 in real-world development scenarios, and whether this consolidation leads to a broader trend of LLM specialization reduction across the industry. The impact on smaller AI companies that might have built tools on top of Codex also warrants attention.