Migrating: GPT-5.6 Sol to GPT-6.1 Sol

If you never chose a model, Codex has already made this move for you: GPT-6.1 Sol replaced GPT-5.6 Sol as the CLI default. This page is for setups that pinned gpt-5.6-sol, or the unsuffixed gpt-5.6 alias that routes to it, and want to follow.

The price case is one-sided

GPT-6.1 Sol is cheaper on every cell of the table — input, cached input, cache writes, output, in both context bands. That holds even against GPT-5.6 Sol's current promotional rates, which OpenAI says are available at least through November 21, 2026; if those rates end, the gap widens rather than closes. The cached-input cell moves furthest, because GPT-6.1 Sol has the deepest cache-read discount of any current model — the row that matters most in long agentic sessions.

What you give up

none reasoning effort. GPT-5.6 Sol accepts everything from none to max; GPT-6.1 Sol starts at low and also rejects minimal. Any request that sets none must change to low — OpenAI's suggested substitute — and will start producing some reasoning tokens, billed as output. On the API, requests that set temperature, top_p or log probabilities need those removed too. GPT-6.1 Sol's rate-limit table also omits the batch-queue column GPT-5.6 Sol publishes.

What stays put

Context window, maximum output, the long-context threshold, and request and token limits per minute.

Before switching

Search for both gpt-5.6-sol and a bare gpt-5.6, since the alias is easy to miss. Run a representative task on both models and compare turns and tokens per finished task, not just per-token price — a newer generation can change how much work a task takes. The effort error page lists the parameter changes in one place.