I don’t really like GPT-5.5…

Theo provides an analysis of OpenAI's latest release, GPT-5.5, highlighting both the advancements and disappointments. He begins by sharing initial enthusiasm but overall disappointment due to the model's limitations. The cost increase of $5 per million tokens in and $30 per million tokens out is noted as significant and justified mainly through the model's efficiency improvements. Theo covers the model's benchmarks, showcasing improvements over previous versions but criticizes the practical aspects of using it. He finds the model capable of understanding complex instructions and performing sophisticated tasks, yet it requires overly detailed prompting and struggles with context retention. The additional sponsor segment for ROR illustrates improved ease and streamlined iOS development. Theo emphasizes the model's inability to honor task intent fully, often requiring new threads to reset contexts. Although it writes high-quality code and performs well in technical tasks, the model's operational quirks and high pricing make it less appealing. Despite these issues, it represents a stepping stone for future advancements, potentially leading to more refined models. Theo ends by suggesting a possible misclassification of the model as GPT-5.5, proposing it be considered distinct enough to warrant its own naming.