GPT-5.5 Review

GPT-5.5 was OpenAI's frontier model from April 24, 2026 until the GPT-5.6 line took over in the summer. It roughly doubled the usable context to a full million tokens over the API, four hundred thousand inside Codex, and accepts files, images and text while returning text. On agentic coding it hit around 88 percent on Terminal-Bench 2.1, which was state of the art at launch and remains genuinely strong.

The problem in September 2026 is value. It still lists at five dollars in and thirty out per million tokens, with a GPT-5.5 Pro tier at thirty and one hundred and eighty, and long context billed at ten and forty five. GPT-5.6 Terra and Luna now deliver similar or better results for less, and Gemini 3.1 Pro undercuts it heavily. Unless you have workflows pinned to this exact model or want the Pro tier's extra headroom, the newer or cheaper options are usually the better buy.

It is a fine choice for teams already standardised on it, for the GPT-5.5 Pro high effort mode on hard problems, and for anyone who values a stable, well understood model over chasing the newest release. For most new projects, start on GPT-5.6 instead.