OpenAI DevDay 2026: agents, teamwork and new ways to use ChatGPT

OpenAI announced more than 20 updates for ChatGPT, Codex and developers. They include the dots assistant, GPT-6.1 Sol and new tools for people to work with AI. Many of the announcements focus on work that continues beyond a single conversation with a model. [1]

Dots: an assistant for longer tasks

OpenAI describes dots as a proactive assistant that can keep working across complex projects and everyday tasks. Its event recap puts particular emphasis on agents that can take on ongoing responsibilities. [1] [2]

This approach is especially interesting where work involves several steps and tracking changes. Rather than having the user explain the task again each time, an assistant would retain its context and continue working within agreed limits.

Permissions become a concrete concern with this kind of use. Which data may the assistant access? What may it do on its own? When must it ask? Drafting a report and sending it to a client require two different permissions.

GPT-6.1 Sol: capability at a lower price

OpenAI also introduced GPT-6.1 Sol. Its official announcement describes a model with near-Astra intelligence for coding, computer use and professional work, at one-fifth of Astra’s standard API input and output token prices. [2]

This is the provider’s comparison, not the result of our own independent testing. It refers to API use in applications and other systems, not to a ChatGPT subscription that costs five times less.

For a company integrating a model into its system, the cost of completing a task correctly matters too. Lower usage prices help if the model maintains sufficient quality on real data without requiring substantially more attempts or manual corrections.

More collaboration in ChatGPT

The announcements include ChatGPT Space, Pages, collaborative slides, teams and shared tasks, as well as @ChatGPT in Slack and Microsoft Teams. OpenAI also announced plugin extensions and automations using MCP events. [1]

These additions give ChatGPT more features for work involving several people. For business users, a useful question is how much copying of information between conversations, documents and colleagues they can eliminate.

When several people prepare the same report, they need shared sources, a clear current version and visibility into changes. Separate AI conversations can help individuals, but do not organize teamwork on their own.

An announcement list does not mean every feature is already included in every subscription. Before adoption, teams will need to check specific availability and member permissions.

Codex expands into the cloud

Codex is gaining more options for cloud and remote use. OpenAI describes reusable development environments that let teams share approved settings and permissions. Other announcements include a refreshed Codex CLI, code review, Codex Security Cloud and an Agents API with computer use. [1]

For development teams, much of the value of these environments lies in repeatability. An agent needs the right dependencies, access to the right files, a way to run tests and clear limits. Even a capable model struggles to complete a task reliably in a poorly configured working environment.

A sensible first test is a bounded task on an existing project: can the agent prepare a change, check it and return a result a developer can review?

DevDay brought enough announcements for several separate trials. We would start with one familiar task and compare the time spent, the correctness of the result and the number of corrections needed. That makes it easier to identify which addition is useful for our work.