GPT-6 Sol vs GPT-5.5

For anyone using Codex with ChatGPT sign-in, this comparison has a deadline. GPT-5.5 retires from ChatGPT, ChatGPT Work and Codex on October 14, 2026, and GPT-6 Sol is the replacement OpenAI names on Plus, Pro, Business, Enterprise and Edu. On the API, GPT-5.5 stays available, so there the comparison is a choice rather than a requirement.

The capacity rows don't move

The context window and maximum output match, both publish the long-context repricing rule, both offer Standard, Batch, Flex and Fast, and the rate-limit table is identical, batch-queue limits included. A workload sized for GPT-5.5's throughput fits GPT-6 Sol without changes.

The price rows all move down

GPT-6 Sol is cheaper on input, cached input and output, and the output gap is the largest. Caching changes shape rather than cost: GPT-5.5 bills written tokens at its ordinary input rate, while GPT-6 Sol has a separate cache-write rate — and that rate is still below GPT-5.5's input rate, so writes cost less too, and reads cost less after them.

What changes in behavior

GPT-6 Sol keeps GPT-5.5's full effort range, none through xhigh with medium as the default, and adds max above it. OpenAI notes that efforts don't map exactly across generations, so test your usual setting and one level lower. The knowledge cutoff moves forward by several months. On the API, migrating from GPT-5.5 means replacing prompt_cache_retention with prompt_cache_options.ttl.

How to decide

In Codex with ChatGPT sign-in the decision is made for you; the only question is whether GPT-6 Sol or the CLI's default, GPT-6.1 Sol, suits your work better. Moving Codex off GPT-5.5 is the checklist.