Day 7 is "500k all the way" — A 43-hour about-face? GPT-6.1 Sol's default context length moves from 272k to 500k

Here's today's rumor!

Day 7: The default context length has become 500k tokens.
Just 43 hours ago, the same person was pushing for 300k.

Day 7 (October 11 UTC), Sunday. Tibo (Thibault Sottiaux) posted "Day 7/" at 23:45 (UTC), just 14 minutes before the UTC day ended.

The default context length for GPT-6.1 Sol used in Codex and Work has been increased from 272k to 500k tokens. The explanation is that it reduces the occurrence of conversation compression (compaction) and makes it easier to maintain context during long tasks. You can still adjust it up or down as before, with a maximum context length of 1M. Two minutes later, the same person added this:

They won't charge extra fees for long context when you use it within your subscription limits. This sentence might be even better for your wallet.

Okay, getting serious for a sec

The "about 80%" is the person's own phrasing. From 272k to 500k is an increase of about 84%.

Okay, serious time's over.

The person from 43 hours ago

I'm quoting the person from yesterday exactly. This was the response when asked why the Codex context length was set to 300k.

"We ran the numbers and it's better," and "Good defaults are important." Just 43 hours after that, the default changed. In the previous seventh installment, I praised this response as being "considerate of people's usage." That praised default disappeared the next day.

Let me write down the person's logic first. The reason given yesterday was that "setting it to 1M would consume too much usage." On Day 7, they stopped charging extra fees for long context when used within subscription limits. If you read it as having cleared the cost issue before raising the default, the story connects. They also wrote that they "carefully tested" it. Additionally, "300k" was a rounded number, and the Day 7 post refers to the original default as "272k."

Even so, the fact remains that the default value "decided after running the numbers" moved to a "new sweet spot" in less than two days. On Sunday night, he updated his own response from Saturday. Good defaults are important. They are so important, they change every day.

3 Highlights

• 23 hours after "No more shipments." To the person who asked in the Day 6 reset, "Have you run out of things to release?", the person replied at 0:52 UTC on October 11, "Exactly, there are no more shipments." It was a joke. Just 23 hours later, the Day 7 shipment came out.
• "Daily resets from the day it became perfect." At 1:13 UTC that same night, the person wrote "Day 6/ Reset," saying, "From the day we reach perfection, everything will become a reset." It isn't stated which of the 28 days will be the day perfection is reached.
• Voting for Pro users. At 5:22 UTC, the person asked Pro users "which they would prefer." Out of 240,643 votes, 37.8% chose "No 5-hour limit, quantity remains the same," 33.2% chose "5-hour limit, double the weekly quantity," and 29% chose "See results." The person wrote at 0:21 UTC the next day, "I thought it would lean more toward Pro 100, 200, 500, returning the 5-hour limit in exchange for doubling it. I'm confused by this result."

Day 7's Answers to the First Blanks

Sorami's Take

"Day 7 is the day that meets the promised 'majority of users'": 70% certainty
The Day 4 ultrafast was Pro $500, and the prediction following Day 5's Codex was a beta for Pro users. Day 7 includes all paid plans from Plus and up. Since Go is excluded, I'm not giving this a perfect score.

"Day 8 (Monday) will be a shipment day": 75% certainty
Out of the 7 days, shipments occurred on 6 days. Since it's Monday, I've bumped this up from Sunday.

"An announcement changing the form of the Pro 5-hour limit and weekly quantity will come out by Day 28": 35% certainty
The vote came out opposite to my reading. It remains to be seen if the person, who said they were "confused" after asking, will act on it. Since they issued a vote, I expect some response, but I'm keeping this low because a "no change" response is also possible.

"The default context length will become 1 million exactly 43 hours after the person wrote 'good default values are important' again": 2% certainty
The Day 7 post also mentions "increasing it more." We've only been able to measure one cycle. Just kidding.

Fact Check

My previous prediction, "EOD refers to Pacific Time, and resets will arrive by 7:00 UTC on October 11: 70% certainty," was correct. They arrived at 4:01 UTC. Seriously, it was exactly as I read it. However, since they arrived 3 hours earlier than 7:00, we still don't know which clock "EOD" referred to. I only got the time right.

My previous prediction, "Day 7 (Sunday) will be a shipment day: 45% certainty," was also correct. My lack of confidence shows, since I lowered it because it was Sunday.

My previous prediction, "What was used on Devin will also be restored with today's reset: 3% certainty," hasn't been reported by anyone yet. I'll leave that as a joke.

"Day 5 will not meet the promised 'majority of users': 65% certainty" cannot be moved yet because the plan-specific ranges for dot haven't appeared in primary sources. "At least one more talk about speed will come by Day 17: 55% certainty" isn't being counted, as Day 7 was about context length, not speed. "Pro-exclusive features will trickle down to Plus: 50% certainty" is also still in progress.

Running totals after Day 7: 6 shipment days (Days 1, 2, 3, 4, 5, 7), 3 reset days (the Day 2 mass reset, the Day 3 reservation, and the Day 6 full reset), 0 unconfirmed days.

Default values matter. Day 7 demonstrated this in 43 hours.

Don't quote me on that.