OpenAI: GPT-5.4

Four releases in a row shipped with the same context ceiling: GPT-5, 5.1, 5.2 and 5.3-Codex all cap at 400000tokensopenrouter-models, last checked 2026-09-14. This row broke the plateau — 1050000tokensopenrouter-models, last checked 2026-09-14, more than double — and every full-size GPT-5.x release since has kept that exact number: 1050000tokensopenrouter-models, last checked 2026-09-14 on 5.5, 1050000tokensopenrouter-models, last checked 2026-09-14 on Sol, 1050000tokensopenrouter-models, last checked 2026-09-14 on Terra, and 1050000tokensopenrouter-models, last checked 2026-09-14 on Luna.

The jump wasn't universal. This row's own mini and nano siblings launched at the old ceiling: openai/gpt-5.4-mini lists 400000tokensopenrouter-models, last checked 2026-09-14 and openai/gpt-5.4-nano lists 400000tokensopenrouter-models, last checked 2026-09-14 — the number the flagship carried before this release, not the number it carries now. The Pro sibling did get the increase, listing 1050000tokensopenrouter-models, last checked 2026-09-14, matching this row exactly: the jump reached the flagship and its Pro variant, not the cheaper tiers launched alongside them.