News
At DevDay on 29 September 2026, OpenAI announced more than 20 new features across ChatGPT, Codex and the API. The common thread: agents that no longer just answer a question but take on ongoing work, operate computers and react to events in other applications – and, if you want, run entirely inside AWS.
The headline feature is Dots. OpenAI describes them as always-on agents that take responsibility for ongoing work, learn what matters to the user and act on their behalf. They are available on Pro and Business Premium in eligible markets; Enterprise, Edu and Healthcare customers get the beta once a workspace administrator enables it.
The Agents API can now operate computers and adds Codex's multi-agent features, tool search, tool calling and context compaction. OpenAI runs the underlying infrastructure.
The Decisions API is built for constrained, real-time decisions: developers define a finite set of questions and possible answers, the Luna model evaluates text or images and returns one of those answers. Typical uses are classification, request routing or picking an agent's next action. It starts as a limited preview, with a broader release planned within days of DevDay.
With Bedrock Managed Agents, powered by OpenAI, OpenAI and Amazon have built managed agents for AWS. They combine the Agents API with AWS-native customization and integration; according to OpenAI, the agents run entirely within AWS.
OpenAI also supports the proposed MCP Events specification: events in connected applications can trigger ChatGPT automations. OpenAI's example: a new task appears on a project board, ChatGPT reads the related documents and prepares a plan on its own.
GPT-6.1 Sol is an upgrade to GPT-6 Sol focused on agentic coding, computer use and professional work. OpenAI says it comes close to GPT-6 Astra at one fifth of Astra's standard token prices.
Ultrafast is a new premium speed tier: up to eight times faster generation in Codex, up to six times faster in the API, and up to 300 tokens per second for GPT-6 Astra. It is available on the new Pro 500 plan (25 times the Plus allowance, according to OpenAI) and on Enterprise; Ultrafast for GPT-6.1 Sol is coming soon.
Besides running locally, Codex now also runs from a phone or in cloud environments that teams can share with preconfigured settings and permissions. The refreshed Codex CLI adds voice control, delegation across multiple tasks, session resumption and worktree support. Also new: a code review workflow in the ChatGPT desktop app that prepares feedback for GitHub pull requests and GitLab merge requests, and Codex Security Cloud, which scans whole GitHub repositories on a schedule and on every new commit and prepares fixes.
Under the name Private Intelligence, OpenAI bundles capabilities for organizations that need stronger protection for their data: Zero Data Retention with Private Safety Processing, an automated safety review in which no OpenAI staff see the content, and Private Inference, which combines confidential computing with verifiable controls. Private Inference is due as a preview in fall 2026.
The rest of the announcements turn ChatGPT into a workplace for teams: ChatGPT Space as shared knowledge for people, ChatGPT and Dots, Pages as a document format for people and agents, collaborative slides with export to PowerPoint and Google Slides, a Meetings plugin for notes and action items, and @ChatGPT in Slack and Microsoft Teams. Plugins get their own place in the sidebar and interactive panels. “Sign in with ChatGPT” lets a ChatGPT subscription be used in 16 partner tools so far, including Notion, Vercel and Devin. Through the OpenAI Marketplace, enterprise customers can put part of their OpenAI commitment toward software from an initial 32 partners such as Figma, Salesforce and CrowdStrike.
For projects that go beyond a chat window, three things stand out. Bedrock Managed Agents lets companies that already run on AWS operate OpenAI agents inside their existing AWS environment. MCP Events gives the Model Context Protocol a way to push: agents no longer have to poll for changes, they get notified. And the Decisions API turns a common building block – a model picking from a fixed list of answers – into a product of its own.
Much of this is still in preview, in beta or tied to expensive plans, though, and every figure on speed, price and performance comes from OpenAI itself. Whether GPT-6.1 Sol really comes close to Astra in your own workflow only a test with your own data will tell.
With taktoro.de and taktoro.com, our new product for companies in Germany is live: time tracking that checks every entry against the German Working Hours Act (ArbZG) before it is saved. Breaks, rest periods, maximum working hours and the record-keeping duty are built in, not booked as add-ons. Developed and hosted in Hamburg.
From 11 to 14 September we're at IBC 2026 in Amsterdam — no booth of our own, but you'll find us hanging around the German Pavilion at the RAI most of the time. We're there to network: catching up with partners, customers and friends from the broadcast and media industry. If you're around too, get in touch.
MJML turns a simple markup language into responsive HTML email, usually with a Node toolchain doing the work behind it. We ported it to PHP end to end: php-mjml/php-mjml needs no Node.js, no second runtime in the deployment. It has rendered this website's emails since February; as of this week it's open under the MIT licence for anyone to use.
