Models#Coding agent#Model inference
GPT-6.1 Sol's default context in Codex and Work rises to 500k tokens
Codex lead Tibo raised GPT-6.1 Sol's default context in Codex and Work from 272k to 500k tokens on October 11, with no long-context fee on subscription usage.
Late on October 11, Thibault “Tibo” Sottiaux, who leads Codex and ChatGPT at OpenAI, announced that GPT-6.1 Sol — the company’s volume-tier model — now defaults to a 500k-token context in Codex and ChatGPT Work, up from 272k. He says the default will keep climbing, and users can override the window up to the model’s 1M ceiling. One provenance note before anything else: as of October 12, the change lives only in that post; the official Codex changelog has not recorded it.
Key points
- 272k to 500k by default: GPT-6.1 Sol’s default context in Codex and ChatGPT Work rose about 84% on October 11, and subscribers need no configuration.
- No surcharge on subscriptions: in a follow-up roughly eight hours later, Tibo confirmed no long-context fees when usage draws from a subscription; the metered API went unaddressed.
- Docs lag the tweet: the changelog’s newest entry is still October 8’s Ultrafast item, so the team lead’s post is currently the only official record.
Background
GPT-6.1 Sol launched at OpenAI’s DevDay on September 29 as the volume tier: near-flagship intelligence at roughly a fifth of GPT-6 Astra’s cost, available from day one in Codex and ChatGPT Work (our release coverage). On October 8 it picked up the Ultrafast tier — 8x speed at 6x price at the top end (previous entry). Tibo is on day 7 of his “28 days of shipping” streak, and the default change is that day’s item. His stated reason is concrete: a larger default window means fewer compactions — the mechanism that squeezes earlier conversation into a summary once the window fills — and every compaction is a chance for a long task to lose its middle.
The facts
The checkable numbers, from the original post and the changelog page (checked October 12):
| Item | Value | Source |
|---|---|---|
| Default context (Codex / Work) | 272k → 500k tokens | Tibo’s post, October 11 |
| Override ceiling | 1M tokens | same post |
| Long-context fee (subscription usage) | none | follow-up, ~8 hours later |
| Stated plans | default will keep rising | same post |
| Official changelog | no entry as of October 12 | newest item is 10-08 Ultrafast |
Two distinctions matter. First, the scope of the fee waiver: the follow-up applies “when drawing usage from the subscription” — Plus and Pro subscribers get the 500k window in Codex and Work without a long-context premium, but whether the metered API carries a long-context surcharge went unanswered. The only third-party traces I could find sit at search-snippet level; no article I could open pins a number down. Second, the arithmetic: 272k to 500k is +83.8%, which the post rounds to about 80%. The announcement itself traveled — roughly 1.2M views and 11k reposts by October 12.
The documentation path is its own signal: Sol’s two earlier Codex and Work milestones — the September 29 launch and October 8’s Ultrafast — each got a formal changelog entry, and this default change is the first to skip one as of my check. Whether it eventually lands there is the simplest test of whether this counts as a shipped release or a trial balloon.
What others say
The pushback arrived quickly. One hot reply asks: “So we’re calling a config edit an actual enhancement now?” A fair hit — 500k is a default, not a new capability, and the model itself is unchanged. Follow-up coverage landed within hours: AIBase under the headline “Long Texts Are Free,” then CryptoBriefing and KuCoin News, all with numbers matching the original post; Chinese outlet IT之家 followed on October 12 via Sina Finance reposts. A competitor angle comes from a DataCamp comparison, visible only as a search snippet because the article would not load: Claude Opus 5.5 charges no long-context surcharge, and Sol’s earlier long-text premium dented its “one-fifth the price” pitch. Worth tracking, but unverifiable against the original text for now.
Our take
For anyone running long sessions in Codex, this is the most practical item of OpenAI’s shipping month: no config change, no model swap, and 228k more default headroom, so mid-task conclusions get summarized away less often. Who can wait: short-task and batch callers, where 272k was rarely the binding constraint and neither pricing nor the model has moved. The pricing signal matters more than the window. Waiving long-context fees inside subscriptions turns long context back into a default experience rather than a premium option, and if it is OpenAI answering rivals that charge no surcharge, heavy users are the winners. Two discounts apply, though. The official docs have not caught up, so treat the changelog’s eventual entry as the binding record and expect possible discrepancies. And as the hot reply says, this is configuration, not capability — reading it as “GPT-6.1 Sol got smarter” is the wrong takeaway.
How to try it
Codex and ChatGPT Work subscribers need to do nothing; the new default should already be live. For sessions that need more room, the post says the context window can be overridden up to 1M — the docs have not been updated with the how, so the tweet is currently the only set of instructions. The quickest check: run a long task into the 500k range and see whether compaction kicks in later than it used to.