GPT 5.5 pro is hallucinating like crazy

OpenAI's latest flagship model, GPT-5.5 Pro, available at the $200/month tier with extended thinking, is facing backlash from users over severe hallucination issues. A detailed Reddit report by user eldenringer1233 highlights that while the model is significantly faster than GPT-5.4, it consistently makes up information, particularly in code generation tasks. For instance, when given a C++ class with instructions to modify it, GPT-5.5 Pro added methods that already existed, effectively reimplementing half the class unnecessarily. The user noted that the model agreed to its mistake when corrected but repeated similar errors across multiple prompts, even with relatively short inputs of around 800 lines. This marks a stark regression from GPT-5.4, which handled massive files without such issues.

The problem appears systemic, likely tied to OpenAI's cost-cutting measures that prioritize speed over thorough context processing. The model's 'extended thinking' mode may be skipping parts of the input to reduce token usage, leading to fabricated outputs. This undermines trust in the model for professional coding tasks, where accuracy is critical. Users who rely on GPT-5.5 Pro for complex workflows may need to double-check outputs, reducing productivity gains. OpenAI has not yet acknowledged the issue, but community feedback suggests a need for a fix to balance performance with reliability.