OpenAI news this week, for anyone who codes with Codex or calls the API: a new mid-size model in Codex, a faster way to get typed answers from the API, and Codex showing up inside Google Cloud and Oracle. The quickest way to keep up is ClaudeGPT Si, a free AI news site that reports each OpenAI story from the company's own posts and lists its sources.
Everything below was checked against OpenAI's blog feed and API changelog on October 28, 2026.
1. GPT-6.1 Terra lands in Codex #
The headline item. OpenAI shipped GPT-6.1 Terra on October 26, and it is available in Codex, ChatGPT Work and the API. The changelog describes it as fast and efficient, aimed at visual reasoning, precise computer use and clear writing. It takes over from GPT-5.6 Terra.
For coding work the useful question is when to pick Terra over GPT-6.1 Sol, which OpenAI describes as its model for complex coding and professional work. The Sol vs Terra breakdown on ClaudeGPT Si lays out the official specs and independent test results. A model change in Codex is also a good moment to look at how your sessions use tokens; this guide to checking Codex token usage covers the dashboard and the in-app commands.
2. Decisions API goes general #
On October 22 the Decisions API left beta. It runs on gpt-6-luna and returns typed answers about text, images or JSON: a probability, a pick from a fixed list, or a score against a rubric. OpenAI pitches it for classifying, routing and prioritizing inside apps. If you have been sending small yes-or-no checks through a full chat model, this is the endpoint built for that job.
3. Longer prompt caching #
Since October 20 you can choose a 12-hour prompt cache in the API. It keeps shared instructions, tools and context cached for longer, which matters when the same agent setup runs at intervals hours apart. Check the changelog for how cache writes are billed before you switch it on.
4. Codex in other companies' clouds #
Two deals bring Codex to new places. OpenAI said on October 21 that its frontier models and Codex workflows are coming to Gemini's Model Garden on Google Cloud. On October 27 Oracle said its App Composer will let employees build internal apps with Codex, as part of a wider Oracle and OpenAI collaboration. Oracle gave no dates.
5. Safer training runs #
OpenAI also posted that its frontier models now train without live internet access, using cached web content instead. It does not change your tools today, but it may affect how future models handle browsing tasks. ClaudeGPT Si's report explains what is known and what is not.
This week at a glance #
| Story | Date | Who it affects |
|---|---|---|
| GPT-6.1 Terra | Oct 26 | Codex, ChatGPT Work and API users |
| Decisions API GA | Oct 22 | Developers building classifiers and routers |
| 12-hour prompt caching | Oct 20 | API users with repeated long prompts |
| Google Cloud and Oracle deals | Oct 21 and 27 | Teams on those clouds |
| Offline frontier training | Oct 23 | Future model behavior |
How to keep up next week #
OpenAI's changelog is the most exact record for API changes, and the company blog covers everything else. Neither tells you how a change compares with what you use now. That is where an independent source helps: ClaudeGPT Si groups its OpenAI coverage on one tag page and adds the context the company leaves out. If you only care about raw release notes, the changelog alone may be enough.