GPT-6.1 Sol and ChatGPT Dots: Sept. 30 dispatch

OpenAI's latest model gives you another way to test the cost of recurring AI work, while ChatGPT gains workers that keep going between conversations. Today's job is to separate three decisions: which model writes your drafts, who monitors unfinished work, and what event starts the next task. Keep customer sending under review.

GPT-6.1 Sol brings a new cost test for recurring work

OpenAI launched GPT-6.1 Sol on September 29, positioning it near its more expensive Astra model for complex work. The DevDay announcement gives you a reason to compare routine document and reporting jobs before continuing to pay Astra rates. The claimed quality is OpenAI's assessment, not a guarantee for your business.

An API is the software connection through which your automation requests an AI answer. Its bill measures tokens, the small pieces of text the model reads and produces. Standard GPT-6.1 Sol prices checked September 30 are below; these rates apply to requests with no more than 272,000 input tokens. Longer requests and optional processing modes have different rates.

Text charged per million tokens	GPT-6.1 Sol
New text supplied	$2
Reused text qualifying for cached pricing	$0.10
Text generated	$10

The part that breaks is the connection, not necessarily the writing. The model documentation says tool calling, which lets the model request actions in connected software, requires the Responses API. The older Chat Completions connection supports this model without those actions. A working paragraph doesn't prove your customer record was updated.

Use the weekly marketing report build as a contained test. If your current cheaper model already passes, leave it running until Sol demonstrates a useful improvement.

ChatGPT Dots takes on work between conversations

OpenAI introduced Dots on September 29: AI agents, meaning software workers that can take actions, with their own online computer and browser. Access is rolling out on eligible Pro and Business Premium accounts; Enterprise access requires administrator enablement. For an owner repeatedly chasing the same project update, this is worth a narrow test. It isn't a reason to hand over customer correspondence.

Follow the first-task setup instructions: open dots in ChatGPT on a desktop browser or the desktop app, then follow the introduction. Connect only the app needed for your first responsibility. You can skip connecting your own computer. Choose a draft campaign plan with known deadlines so you can judge whether the worker spots the right missing decisions.

Copy this into your first conversation, replacing the document description with your actual plan:

Track the campaign plan I shared. Identify deadlines and unanswered questions. Draft a short update for my review. Do not send messages, publish content, or change the source plan. Ask me before expanding the task. Confirm which sources you can access.

In the desktop app, open your dot's profile and select Activity to inspect delegated work and questions awaiting your input. Ending the conversation doesn't necessarily stop assigned work, so explicitly cancel responsibilities you no longer want. Keep the content approval checklist attached to the process: checking a draft still needs a named person.

ChatGPT plugins gain interfaces and event triggers

OpenAI's September 29 plugin expansion lets developers add sidebar tools and interactive panels inside ChatGPT, alongside support for automations triggered by activity in another app. A plugin is an add-on that connects software and instructions to ChatGPT. For campaign reviews, the useful change is having a new comment start a draft revision without someone remembering to ask.

The event integration documentation describes MCP Events, a shared format for connected software to announce changes. It requires MCP 2.0 and webhook delivery: sending data to a receiving web address. Repeatedly checking for updates, called polling, isn't supported by this integration. This needs implementation work; installing an existing plugin doesn't establish that it supports events.

Ask your plugin maintainer to demonstrate one document subscription, meaning a saved instruction to watch that document. Then add a test comment and confirm that ChatGPT receives it and drafts a response. Add another comment to an unrelated document; it should stay outside the subscription. Finally, cancel monitoring and check that deliveries stop. Use the webhook setup guide to understand the sending-and-receiving pattern, but don't treat its Zapier steps as this integration's instructions.

On the bench

• GPT-6.1 Sol Ultrafast: Announced as coming soon. Save a slow task for testing before budgeting for extra speed.
• ChatGPT collaborative slides: Due in the coming weeks. Keep a presentation with awkward formatting ready to check export quality.

Get the next build from the workflow collection.