OpenAI’s 28 Days of Shipping: Which Updates Actually Matter?

Checked October 19, 2026, 10:43 SGT. Latest verified announcement: Day 14. Timeline dates use Pacific Time; Days 15–28 remain pending.

OpenAI’s 28 Days of Shipping is turning daily product announcements into a practical question: can you finish more useful work with fewer interruptions? For people moving between research, campaign planning, writing, and code, the most valuable update may be a quieter improvement that keeps a task moving.

So far, the strongest themes are faster responses, more room for long tasks, and better use of your allowance. Start with the complete timeline below, then see which changes deserve your attention. Lasting product improvements are distinguished from temporary usage relief.

What Is OpenAI’s 28 Days of Shipping?

In his original commitment, OpenAI’s Tibo promised a relevant improvement for most Codex or Work users each day for 28 days, with a full usage reset as the alternative. Day 1 followed on October 5, Pacific Time.

That wording matters: this is an ongoing commitment, not 28 completed feature launches. A reset restores usage allowance; it does not add a permanent capability. Availability also depends on the specific product, plan, and rollout.

OpenAI 28 Days of Shipping: Complete Timeline So Far

Every announced day is included below, newest first. Links point to original announcements. Pending days are not treated as missed commitments.

Day	Date, 2026 (PT)	Announcement and Practical Meaning
14	October 18	Edit earlier messages in desktop threads. The post does not specify plan or operating-system restrictions.
13	October 17	Usage-limit reset announced, with rollout scheduled by the end of the Pacific day. Check your account for the allowance applied.
12	October 16	More effective usage through context reuse across subagents, longer caching, and improved routing. The thread also highlights an Astra Pro upgrade and Home Screen dots.
11	October 15	Capacity improvements reduce model-capacity interruptions; affected requests should eventually resume automatically. No recovery-time guarantee was given.
10	October 14	Usage reset announced, followed by rollout timing. This is temporary usage relief.
9	October 13	Desktop storage optimizations expected to reduce stored application state by 60%, rather than total computer storage.
8	October 12	Global usage reset announced. Confirm the result in your account.
7	October 11	500,000-token default context for GPT-6.1 Sol in Codex and Work; configurable up to 1 million. Subscription usage has no separate long-context fee.
6	October 10	Global reset announced for the end of the day.
5	October 9	Composer predictions beta for Codex desktop Pro users, plus creating and texting dots on mobile.
4	October 8	Instant steering and Sol Ultrafast, combining quicker direction changes with a faster eligible model mode.
3	October 7	GPT-6 in Chat and a banked reset announced for paid accounts. A banked reset differs from an automatic refresh.
2	October 6	Included Auto-review, simpler API tiers, integrated meeting notes, and the Decisions API: several updates serving different audiences.
1	October 5	Default speed improvement of approximately 50% for GPT-6 Astra and GPT-6.1 Sol across subscription products and eligible Sign in With ChatGPT partners.

Days 15–28: Pending as of the verification time above. No future releases are assumed.

The Updates That Matter Most

Faster Iteration and Easier Course Correction

Day 1 increased default model speed, while Day 4 introduced faster steering and GPT-6.1 Sol Ultrafast. Steering is useful when a brief changes mid-task: you can redirect work sooner instead of waiting for an unwanted answer to finish.

For a marketer refining a campaign or a developer adjusting a prototype, the likely benefit is a shorter feedback loop. These announcements do not establish that every project finishes proportionally faster. Planning, tools, and review still take time.

Ultrafast also has a narrower audience. Official speed guidance lists Pro $500 and eligible Enterprise and Edu access; Enterprise owners must enable it. As of October 19, included subscription limits are consumed at eight times the Standard rate; purchased credits and Enterprise pay-as-you-go usage are billed at six times the Standard rate. Faster output comes with a usage tradeoff.

More Continuity for Substantial Tasks

The Day 7 announcement raised GPT-6.1 Sol’s default context in Codex and Work from 272,000 to 500,000 tokens. Context is the material a model can work with at once. More room can help with a long research brief or a project containing many files, with fewer context-compaction steps.

Day 11 addressed capacity interruptions, and Day 12 improved context reuse and caching. Together, these changes target the friction of keeping substantial work moving. The caching announcement gives no percentage increase in effective usage, so there is no reliable universal productivity multiplier to quote.

Small Changes That Reduce Everyday Friction

Composer predictions suggest a possible follow-up message, while Day 14 adds editing of earlier messages in desktop threads. These are modest improvements, but frequent users may encounter them more often than a specialized model mode.

Composer predictions remain a beta for eligible adult users on personal Pro plans in supported regions. Suggestions become editable drafts; accepting one does not automatically send it.

What Should You Try First?

Start with a task you already understand: revise a campaign brief, compare research documents, or improve a small website. Judge the update by completed output and how much correction it needs.

If interruptions are your main problem, watch task continuity and capacity recovery. If your work involves substantial background material, evaluate the larger context. If you repeatedly hit usage limits, check your allowance and reset availability before starting another long run.

For creators, try a concrete handoff: turn research into a campaign brief, refine the script through several rounds, then carry the approved direction into image or video production. Measure the revisions needed before that handoff. This reveals more than testing how quickly a model produces its first paragraph.

OpenAI 28 Days of Shipping is most useful as a guide to changes you can actually use. Prioritize improvements that remove a recurring obstacle, verify access, and keep the timeline handy as the remaining days arrive.