gpt-5.1-codex-mini

GPT-5.1 Codex is a specialized version optimized for long-term, agent-enabled coding tasks, built on the foundation of GPT-5.1.It responds to simple tasks up to twice as fast as GPT-5,enabling continuous dialogue and autonomous execution in Codex or similar environments, with a context window of 400,000 tokens and supporting a maximum output limit of 128,000 tokens.In the SWE-bench Verified benchmark, GPT-5.1 Codex improved its score from 72.8% with GPT-5 to 76.3%;this model focuses specifically on software engineering tasks, such as translating requirements into code, feature development, debugging, refactoring, and code review,offering eight distinct conversational styles—including default, professional, friendly, and others—to accommodate a wide range of scenarios, from rigorous code reviews to creative programming. Users can also fine-tune parameters like "simplicity" and "enthusiasm" to tailor the model into a personalized programming assistant.Compared to general-purpose models, it boasts greater adaptability and alignment with developer instructions, making it ideal for extended use within IDEs/CLIs and the Codex toolchain. It’s particularly well-suited for ongoing development of large-scale codebases, cross-file refactoring, long-form collaborative pair programming, and engineering tasks that require advanced tool or agent orchestration.

Official Guide:https://cookbook.openai.com/examples/gpt-5-codex_prompting_guide