Migrating: GPT-5.6 Terra to GPT-6.1 Terra

OpenAI's GPT-6 guide names this migration directly: if you're moving from GPT-5.6 Terra, it says to compare both models on representative images and workflows at the same reasoning effort. On price the case is already made — the table above shows the newer Terra cheaper on every Standard cell — so what's left to check is behavior.

What gets cheaper

Input, cached input, cache writes and output, in both context bands and at every tier both publish. The cached-input row moves furthest: GPT-6.1 Terra bills a cached read at 0.05x its input rate instead of 0.1x, on a lower input rate to begin with, so cache-heavy agentic sessions gain the most.

What carries over

The context window and maximum output, the effort range from none to max with medium as the default, the rate-limit table — request, token and batch-queue limits tier for tier — and the service tiers: Standard, Batch, Flex and Fast on both, Ultrafast on neither. Cache writes keep their own rate, and the long-context rule still reprices the whole request past 272K input tokens; the newer page's sentence names the cache rates in that repricing explicitly, which the older one's published bands already did.

What changes

The knowledge cutoff moves forward by several months. Tool calls belong on the Responses API: OpenAI's advice for this move is that an application calling tools through Chat Completions should migrate those requests. And GPT-6.1 Terra supports only US and EU data residency, so a workload pinned to another region should check before switching.

Inside Codex

GPT-5.6 Terra stays available, and OpenAI says its credit rates are unchanged; GPT-6.1 Terra's are lower. The newer model needs CLI 0.159.0 or later — select it with codex -m gpt-6.1-terra or /model.

Before switching

Run the same images and workflows on both at your current effort level, as OpenAI suggests, and compare finished results rather than single answers. If the output holds up, every row of the bill gets smaller.