OpenAI discontinues separate Codex line, merges coding capabilities into GPT-5.5

OpenAI will not release a separate GPT-5.5-Codex model, according to Romain Huet, who confirmed the company has unified its coding-specialized model line with its main GPT models.

Codex merged since GPT-5.4

"Since GPT-5.4, we've unified Codex and the main model into a single system, so there's no separate coding line anymore," Huet stated. The Codex line, previously OpenAI's specialized offering for code generation tasks, has been absorbed into the primary GPT model family.

GPT-5.5 represents the continuation of this unified approach, with Huet claiming "strong gains in agentic coding, computer use, and any task on a computer." Specific benchmark scores or quantitative performance improvements were not disclosed.

What this means

The consolidation eliminates the need for developers to choose between general-purpose and code-specialized models. This architectural shift suggests OpenAI believes a single multimodal model can match or exceed the performance of task-specific variants, potentially simplifying their product lineup and reducing infrastructure overhead.

The emphasis on "agentic coding" and "computer use" signals OpenAI's focus on models that can autonomously navigate software environments and execute multi-step programming tasks, aligning with the broader industry trend toward AI agents. However, without published benchmarks comparing GPT-5.5 to the legacy Codex line on standard coding evaluations like HumanEval or MBPP, the actual performance gains remain unverified.

Pricing and context window specifications for GPT-5.5 have not been announced.