28 days of Quality of Life improvements and/or a full Reset

It looks like the OpenAI team decided that the flood of DevDay announcements was not enough. :grinning_face_with_smiling_eyes:

Starting today, there will be one clear improvement each day for most of all Codex or Work users. On days without one, everyone gets a full reset instead and sometimes we get both.

We’ll keep this topic updated each day.


Click on any announcement to jump straight to the relevant post.

Day 1: GPT-6-Astra and GPT-6.1-Sol performance upgrade

Day 2.1: Auto-Review is now free

Day 2.2: Simplified API rate-limit qualification

Day 2.3: Meetings plugin for notetaking

Day 2.4: Decisions API is in public beta

Day 2.5: Rate Limits reset for everyone

Day 3: GPT‑6 and Intelligent UI for everyone

Day 3.1: A banked reset for all paid accounts to celebrate 40 million Codes and Work users

Day 4.1: Ultrafast is rolling out today for GPT-6.1 Sol in the API, Codex, and ChatGPT Work

Day 4.2: Near instant steering of conversations

Day 5.1: Updates for Dots

Day 5.2: Composer predictions in Codex for Pro users.

Day 6: Reset

Day 7: Default context length for GPT 6.1-Sol is now 500k tokens

Day 8: Reset

Day 9: Codex Desktop Storage Optimizations

Day 10: Reset

Day 11: More compute, more capacity, fewer interruptions

Day 12.1: Upgraded GPT-6 Astra Pro

Day 12.2: Increased effective usage in Codex and ChatGPT Work

Day 13: Reset

Day 14: It’s now possible to edit any user message in Codex, not just the last one.

Day 15.1: Video: Cooking with Codex

Day 15.2: It’s a full reset

Day 16.1: You can now choose 12-hour prompt caching in the API.

Day 16.2: 12hr Cache is now enabled in Codex and ChatGPT Work

Day 17: It’s a reset

Day 18: It’s a reset

Day 19: a set of small quality of life updates

Day 20: It’s a Reset

Day 21: It’s a Reset.

Day 22: Release of GPT-6.1 Terra: Fast, capable, low prices

Day 23: The iCloud and calendar plug-ins don’t count so we get a reset

Day 24: You can now text your Dot via iMessage or RCS

Day 25: Double usage limits for Pro and Business Premium

Day 26: Stay signed in to and easily switch between two accounts

Day 27: It’s a banked reset

Day 28: It’s a banked reset


Day 1:

GPT-6 Astra and GPT-6.1 Sol should now feel around 50% faster by default when used through your subscription.

This applies across OpenAI products and partners that support Sign in with ChatGPT, including OpenCode, Pi, Amp, and Devin.

There is nothing you need to change. You should notice the improvement within the next two hours.

Source

is this code for “they discovered a performance improvement”?

Yes. That’s a performance improvement, but we don’t know how they did it just yet.

Here, @reach_vb says it’s an inference improvement.

Well that’s their end of year bonus paid for :wink:

This rebroadcast of promotional tweets for end-user subscription products seems like something belonging in the “ChatGPT” forum category.

No changes to API speed, lest you need to re-run evals against the symptoms of the “optimization”?

Looks like Improvement: ChatGPT put a “full width” option in the appearance settings.
But not an improvement: that option doesn’t make the chat contents wider to 100% like a nice Tampermonkey script might, so you can see all the code: it just auto-hides the chat history sidebar.

An improvement on a bad improvement: Now turn off the auto-rich-formatting of what you type while you type as an offered option in settings, also in appearance → advanced.

I really hope the UI interactive feature can come back.
A few days ago ChatGPT was able to create interactive study blocks directly in the chat, with answer boxes, a “Submit answers” button, scores, feedback for each answer and then the next questions based on the mistakes.
For studying it was honestly much better than plain text. It made active recall feel much more natural and useful.
It would be great if this could be restored as a normal feature!!!

Many people I know will never go to Twitter, and we are offering this as a service so everyone can keep up with what’s going on. Also, it’s fun.

Codex is a developer product that can be accessed at a steep discount compared to API pricing when using a subscription.

I’m not sure I would choose to have the team update the interface instead of improving inference speed. The latter is something I cannot do myself, while interface changes can be handled with something like a Tampermonkey script.

Does yours still work? Link?

OpenAI has changed the targeted HTML element tree, so currently I have reversion of much of the styling I’ve also had to expectedly re-tweak in the past, including width. I haven’t made personal rule-writing updates of script for over a month, and stopped sharing long ago because of expectation a Tampermonkey script could break at any push of the underlying page structure.

With the “wide” option in settings OpenAI offers, there is no actual wider content, and clipped code in an overflow surrounded by wasted space is the symptom:

It is sufficient to say the presentation is not ideal when “bubble vision” (wanting to look like iPhone texting with un-named roles) returns to break even a single line of input, with only a left flex position and no radius working in my !important CSS (but at least is not a misalignment on the right):

Appreciation of a more work-oriented UI need would be a valuable change.

One thing that is certain when planning is the daily uncertainty in the value and availability of a stable ChatGPT surface and shape to which one might be getting acclimated - an acclimation driven by marketing campaign.

Not entirely convinced the speed-up is having a positive effect on 6.1 Sol:

RANK MODEL EFFORT TASK RESULT TIME TOKENS API EQ
#1 GPT-6-Sol HIGH DIGBENCH P-1 WIN 8m45s 3.7M ~$1.1648
#1 GPT-6-Sol HIGH DIGBENCH P-2 WIN 23m19s 19.2M ~$4.8133
#2 GPT-6.1-Sol HIGH DIGBENCH P-1 WIN 34m36s 45.2M ~$5.3291
#2 GPT-6.1-Sol HIGH DIGBENCH P-2 WIN 30m56s 17.5M ~$2.6719

These were my last results, and this would suggest there has been some service degradation for 6.1, but an improvement for 6.0

I’ll rerun this in the morning, perhaps it’s a combo of US hours and a longer than expected rollout …

I’d like the DOT phone call feature to not autolock my iphone and hang up on me … normal chat voice doesn’t do that. It’s an app bug.

I just want to know when the chat mode for GPT-6 will be available.:grinning_face:

Tibo’s statements are always vague and difficult to understand (as we’ve already experienced with the reset time), so I think it’s great to have a place where we can repost things like this and discuss them
I welcome the increase to 150%

Day 2:

Auto-review is now free for everyone signed in with a ChatGPT account. You can enable it under Settings → Permissions → Auto-review.

For longer tasks, approving every action quickly becomes tiring unless you have spent time configuring individual rules. Auto-review adds a second agent that reviews the primary agent’s actions for you.

Its job is limited to stopping high-risk actions and anything that does not match your original request. It also does not count against your plan usage.

Source

OK here’s a run this morning whilst US sleeps and hopefully safely post rollout of the performance upgrade.

This is looking a lot more healthy:

RANK MODEL EFFORT TASK RESULT TIME TOKENS API EQ
#1 GPT-6.1-Sol HIGH DIGBENCH P-1 WIN 9m40s 6.4M ~$1.1066
#1 GPT-6.1-Sol HIGH DIGBENCH P-2 WIN 24m48s 18.6M ~$2.4581
#2 GPT-6-Sol HIGH DIGBENCH P-1 WIN 9m01s 5.2M ~$1.5423
#2 GPT-6-Sol HIGH DIGBENCH P-2 WIN 20m30s 13M ~$3.4637

and for 6.1 beats both prior runs

Hey Tibo — here’s something to fix for your next improvement of the day :slightly_smiling_face:

The Codex VS Code queue bug still reproduces in 26.930.61225. I’ve released my fourth community patch, with reproducible tests and detailed results.

Unfortunately, I can’t post links here. On GitHub, open OpenAI’s Codex repository → Discussions → #50980.

The discussion title is:
“How many more Codex VS Code releases will ship the same queue bug? Fourth community patch is out.”

Posted by IveSan83. Bring popcorn. :popcorn:

Sharing a friendly twist on Tibo’s 28-day challenge :slightly_smiling_face:

What if resets became a celebration of improvements, not just the fallback?

When a meaningful improvement ships for most users, celebrate it with a global usage reset—so even users with depleted quotas can explore it properly and share concrete feedback.

On days without a major improvement, offer a banked reset, to activate when it fits each user’s workflow.

The idea is to connect every improvement with an opportunity to actually experience it: the team ships, users test, and the feedback loop gets stronger.

I shared the suggestion with Tibo on" X:

Would this help your workflow? I’d love to hear both the benefits and the trade-offs. If the idea resonates, feel free to support or share the original post on X so the discussion is easier to find.