GPT-6.1 Sol vs GPT-5.6 Sol

This is the old Codex CLI default against the new one. GPT-6.1 Sol replaced GPT-5.6 Sol as the model the CLI runs when you don't choose, so any setup that never pinned a model has already made this switch — and any setup that pinned gpt-5.6-sol, or the unsuffixed gpt-5.6 alias, hasn't.

What the newer model wins

Price, on every cell. GPT-6.1 Sol is cheaper per input token, per cached input token, per cache write and per output token — and that comparison is against GPT-5.6 Sol's promotional rates, which OpenAI says are available at least through November 21, 2026. Its knowledge cutoff is a couple of months newer. And its cached-input discount is the deepest of any current model, which compounds in long agentic sessions.

What the older model keeps

The none effort level. GPT-5.6 Sol accepts everything from none to max; GPT-6.1 Sol starts at low. A workload that switches reasoning off for speed can't carry that setting across. GPT-5.6 Sol also publishes a batch-queue limit per usage tier, which GPT-6.1 Sol's table omits. And it has a track record on your workload, which is worth something if your prompts are tuned to it.

What stays the same

The context window, maximum output, long-context threshold and request and token limits per minute all match. Both offer Standard, Batch, Flex and Fast.

How to decide

Unless you depend on none effort, the newer default is cheaper on every row and newer on recency. Test a representative task on both before moving a tuned workload, and see the migration page for what to change in config.