GPT-5.1-Codex-Max

GPT-5.1-Codex-Max reached the Responses API on 4 December 2025, three weeks after the rest of the GPT-5.1 Codex line. OpenAI calls it purpose-built for agentic coding and optimized for long running tasks, and in the changelog goes further: its most intelligent coding model for long-horizon, agentic work. Set its card next to GPT-5.1-Codex and the interesting part is how little changes. Same 400,000-token context window, same 128,000-token output ceiling, same 30 September 2024 knowledge cutoff, same text-and-image input, same tools - function calling and web search. Even the price is identical to the digit: 1.25 USD per million input tokens, 0.125 USD cached, 10 USD per million output. Whatever makes the Max version better at multi-hour runs is not a specification difference and OpenAI does not quantify it on the model card; there are no benchmark figures there at all. The practical reading is that Max is not an upsell tier but a second default for a different shape of task: if a job runs long enough that the agent has to keep its own thread across many steps, this is the model OpenAI points at, at no extra cost per token. As elsewhere in the line, it is Responses API only, with no Chat Completions, batch or fine-tuning route. The model is no longer available. Access was shut down on 23 July 2026, seven and a half months after release, in the group withdrawal announced on 22 April 2026; the substitute named in the table is GPT-5.6 Sol. Its short life is worth noting next to OpenAI's own commitment: specialised variants of generally available models, Codex versions among them, carry only a three-month notice period, half the six months promised for mainline models.

▸Videos

No videos yet.

▸Reviews