On 11 October 2026 at 23:45 UTC, Codex lead Tibo wrote that the default context length for GPT-6.1 Sol, when used through Codex and Work, is now 500k, up from 272k. He called that about an 80% increase. The post says the higher default should reduce how often long tasks compact, and should help work that has to keep detail over time. He says 500k was carefully tested and is the new sweet spot for most users, and that the default will go higher later.
Tibo wrote “Codex & Work.” OpenAI’s changelog and pricing pages use the name ChatGPT Work for that product surface.
The change is a default. Tibo says people can still override it higher or lower, and he puts GPT-6.1 Sol’s maximum context length at 1M. The model page lists a 1,050,000-token context window, a maximum input of 922,000 tokens, and a maximum output of 128,000 tokens. Those are the documented model limits. They are not a measurement of the new Codex default.
What the API still charges past 272K
The model page says prompts with more than 272K input tokens are priced at 2x input and cache rates and 1.5x output for the full request. That sentence is API pricing. Standard rates on the same page are $2 per million input tokens and $10 per million output tokens, before that multiplier.
AICoder writes that subscription usage carries no long-context fees, and it attributes the sentence to Tibo. The Day 7 post does not say that. The Codex pricing page says API token prices are separate from subscription usage, and that those API prices should not be used to estimate included tasks. It does not state a long-context multiplier for included plan usage, and it does not say the multiplier is waived.
List-price context for the model family is in GPT-6.1 Sol beside Astra and GPT-6 Sol and Luna.
Overrides in the config file
The configuration reference says user-level configuration lives in ~/.codex/config.toml. It documents model_context_window as the context window, in tokens, available to the active model, and model_auto_compact_token_limit as the token threshold that triggers automatic history compaction. The page says an unset compaction limit uses model defaults. AICoder names those two keys as the override for the new default. The reference describes the keys. It does not publish 500k as the shipped number.
GPT-5.5 leaves ChatGPT and Codex on 14 October
A changelog entry dated 14 September 2026 says that on 14 October 2026 GPT-5.5 retires from ChatGPT, ChatGPT Work, and Codex on all plans, including consumer, Business, Enterprise, and Edu. The retirement does not apply to the OpenAI API. The pricing page repeats the date and the API exception. AICoder’s summary names ChatGPT and Codex. The changelog also names ChatGPT Work.
The day before
On 10 October, Tibo’s Day 6 post is the text “Day 6/ Reset.” A reply one minute later says “Global reset by EOD. May the tokens do good things for you.” On 11 October at 04:01 UTC he posted “Put the reset in the bag. All propagated.” That is the Day 6 record in the posts themselves: a reset he described as global, due by the end of the day, and later called propagated.
No score attached to the default
The Day 7 post does not include a benchmark, a latency figure, or a comparison suite for the 500k default. “Carefully tested” and “sweet spot” are Tibo’s account of OpenAI’s own check. Separate Codex notes on the site cover reusable cloud environments and the Windows sandbox.
The Campfire
No commentsNobody has pulled up a log by this one yet. Be the first to say what you make of it.
Held for the desk. It appears after a look.