OpenAI releases GPT-5.5, a model designed for complex reasoning and multi-tool workflows

OpenAI released GPT-5.5 on April 23, 2026, describing the model as purpose-built for complex, real-world applications spanning code writing, online research, information analysis, document creation, and multi-tool task execution. According to the company's system card, the model demonstrates improved task comprehension compared to earlier versions, requiring less explicit user direction while making more effective use of external tools.

Before deployment, GPT-5.5 underwent OpenAI's standard predeployment safety evaluation suite and Preparedness Framework. The evaluation included targeted red-teaming focused on advanced cybersecurity and biology capabilities. OpenAI collected real-world feedback from approximately 200 early-access partners prior to release.

The company states it is releasing GPT-5.5 with safeguards designed to reduce misuse while preserving legitimate uses of advanced capabilities. OpenAI describes GPT-5.5 Pro, a variant using parallel test-time compute, as relying on the same underlying model, though the company notes it conducts separate evaluations for that setting in cases where the computational approach could materially affect relevant risks or appropriate safeguard posture.