GPT‑5.5 in Copilot

OpenAI released GPT‑5.5 on April 23, and GitHub started rolling it out to GitHub Copilot one day later. (openai.com) (github.blog) OpenAI said GPT‑5.5 is built for “real work” across coding, research, data analysis, documents, spreadsheets, and software tools, with the model planning steps and checking its own work. The company said GPT‑5.5 matches GPT‑5.4’s per-token latency while using fewer tokens on Codex tasks. (openai.com 1) (openai.com 2) GitHub said its early testing showed the model’s strongest results on “complex, multi-step agentic coding” and on real-world coding problems previous GPT models could not solve. The Copilot changelog lists GPT‑5.5 as generally available and rolling out. (github.blog) Agentic coding means giving a model a larger programming job — like tracing bugs across files, editing code, running tools, and iterating until the task is done — instead of asking for one snippet at a time. OpenAI’s system card says GPT‑5.5 asks for less guidance, uses tools more effectively, and keeps going until a task is finished. (openai.com) That matters for Copilot because GitHub says those workflows have changed the product’s compute demands. On April 20, GitHub said it was pausing new sign-ups for Pro, Pro+, and Student plans, tightening usage limits, and adjusting model availability to protect reliability. (github.blog 1) (github.blog 2) GitHub said the new limits include session caps and weekly token usage limits, and that premium-model access can be disabled after users hit those thresholds. The company said Copilot then falls back to automatic model selection with less resource-intensive models. (github.blog) (windowsreport.com) OpenAI is also pitching GPT‑5.5 beyond Copilot. The company said the model rolled out on April 23 to ChatGPT Plus, Pro, Business, and Enterprise users, while GPT‑5.5 Pro rolled out to Pro, Business, and Enterprise users, and both became available in the application programming interface on April 24. (openai.com) Early hands-on testing has been positive but not unqualified. ZDNET’s David Gewirtz gave GPT‑5.5 a 93 out of 100 in a 10-round test and said it sometimes ignored simple directions because of “exuberance.” (zdnet.com) OpenAI said it evaluated GPT‑5.5 with internal and external red-teamers and collected feedback from nearly 200 early-access partners before release. The immediate test for GitHub is whether a stronger coding model can stay fast and affordable inside a Copilot service that is already rationing access. (openai.com) (github.blog)