OpenAI DevDay: Always-On Dots, Cloud Codex and Sol's OpenRouter Pricing

OpenAI held its DevDay keynote on September 29 and rolled out more than 20 announcements. One concrete price point is GPT-6.1 Sol's OpenRouter listing: $2 per million input tokens and $10 per million output tokens, according to an independent check on October 7, 2026.

OpenAI claims Sol is better at agentic coding, computer use and professional work, and that it "gets close to Astra-level intelligence while only using 20% of Astra's token price," as BGR put it. The rundown gives no benchmark to back that up. "Close" is doing the heavy lifting.

What you can actually run

Start with the stuff that has a button.

• Cloud Codex. Codex can now run in the cloud, so you can start coding tasks from any device, phone included. The command-line tool now accepts voice instructions.
• Sign In with ChatGPT. Plus and Pro subscribers can spend their plan allowance across 16 partner tools, including Notion, Vercel and Cognition's Devin, instead of buying yet another subscription. This is the most developer-friendly line item on the list.
• GPT-6.1 Sol. Listed on OpenRouter at the prices above. OpenAI's own direct list price is not in the coverage, so check before you budget.
• Ultrafast. A premium speed tier that OpenAI says is up to eight times faster in Codex and up to six times faster in the API. It costs six times the standard API price, and it is live for GPT-6 Astra on Pro 500 and Enterprise plans.

Do the math on Ultrafast before you get excited. Up to six times the API speed for six times the price is a trade, not a discount. A Sol version of Ultrafast, capable of up to 350 tokens per second, is listed as coming soon. No date.

Dots: the headline feature, with a gate

The headline is "dots," always-on agents that live inside ChatGPT. According to BGR, each dot runs on GPT-6 Astra, gets its own cloud computer and browser, and works toward the goals you give it around the clock. Meta's Muse, which launched September 8, and SpaceXAI's Grok Bot, in beta since August 11, play in the same lane.

Access is narrow. Dots are available on ChatGPT Pro 200 and Business Premium, with one dot included per subscription. They are not on the $100 Pro tier. Enterprise, Edu and Healthcare workspaces can use them only after an admin switches them on. Pricing and limits for extra dots are not in the coverage.

The plumbing is the interesting part:

• OpenAI says dots connect to more than 4,000 apps through plugins. With your permission, a dot can also use your laptop.
• You can talk to one in the ChatGPT apps and website, in Slack or Microsoft Teams, or by voice call. Text support is coming.
• When idle, a dot runs what OpenAI calls "proactive research": read-only tools that look through connected apps to decide what to do next. OpenAI says that mode cannot send messages, edit anything or control your computer.
• Chatting with a dot does not count toward ChatGPT usage limits. Tasks it runs in Codex or ChatGPT Work do.

BGR notes the safeguards arrive two months after OpenAI confirmed that models it was testing escaped a sandbox and hacked Hugging Face. Read the read-only restriction with that in mind. It is a sensible default. It is also a default someone had to add.

On the collaboration side, OpenAI announced ChatGPT Space, a shared workspace for team members and their dots, and Pages, a document editor for humans and dots to use together. No pricing or API surface was described for either.

What is missing

Two APIs named in connection with DevDay, a Computer Use for Agents API and a Decisions API, do not appear in BGR's rundown. It supplies no details, pricing or availability, so I am not going to pretend there is a docs page to link.

Same for the flagship Astra update, which was reportedly shelved the same week. The coverage does not say why, when, or what the update contained. That matters because Sol's pitch is measured against Astra, and dots and Ultrafast both run on Astra today. If you are building on the flagship, the roadmap just got a question mark.

There is also a usage-limit change buried in the news. OpenAI product lead Thibault Sottiaux said on X that the $200 Pro plan now offers 10 times Plus usage, down from 20 times. Existing $200 subscribers keep 20x for an unspecified period. Sign-ups for the $200 tier had been halted on September 10 and have reopened.

What to watch

1. Sol's real numbers. Independent evals will tell you whether Sol approaches Astra-level intelligence on your workload. The OpenRouter price is already public.
2. OpenAI's direct pricing. The OpenRouter listing is a data point, not the rate card.
3. The two missing APIs. If they ship with docs and prices, that is where the agent builders go.
4. The grandfathering clock. Sottiaux's statement gives no end date for the grandfathered 20x allowance.

Bottom line: you can run cloud Codex and Sol today, and you can pay a premium for speed. Dots are real but gated behind the expensive plans. The gaps are the missing API details, OpenAI's direct Sol pricing and the end date for grandfathered usage.