OpenAI fires back at Anthropic with dots at DevDay 2026

Sam Altman has walked onto the DevDay 2026 stage with more than 20 launches to show off, and after the last few weeks, he needs every one of them.

It has been a rough run for OpenAI. Anthropic’s Claude Opus 5.5 beat GPT-6 Astra on most benchmarks last week, and Sonnet 5.5 landed on Monday to make matters worse. The Wall Street Journal also reported that OpenAI scrapped GPT-6.1 Astra after it showed a higher tendency to deceive and ignore user guardrails. Meanwhile, Meta’s Muse agent is breaking adoption records. Ouch.

There is one bright spot. Axios reports that OpenAI’s annual recurring revenue is close to $68 billion, up over 70 percent since the start of Q3 2026. For context, Anthropic was sitting near $65 billion in August. So the money is fine. The mood is another matter.

Dots are OpenAI’s answer to Meta Muse

The headline act is Dots, personal AI agents that stay on all the time and run on GPT-6 Astra. Each Dot gets its own cloud computer and browser, and it can connect to over 4,000 apps through plugins, including Slack and Teams. You can message or call it in ChatGPT, and it will ping you when it needs a decision. Texting is coming soon. OpenAI also says Dots can use saved passwords to sign in to websites without exposing them to the model, which is a sensible touch.

Dots roll out today to Pro and Business Premium users, except in the EEA, Switzerland and the UK. Enterprise users get a beta once their workspace admin switches it on. Your first Dot comes with your plan at no extra cost, and for the next month, chatting with it won’t count against your allowances. OpenAI is also bringing Dots to Microsoft Agent 365 and hosting them on AWS with Amazon.

The demo did hit a snag, though. The agent failed to respond on stage. Awkward. Dots also look built for businesses first, and with Muse free up to a limit, I reckon Meta keeps the consumer crowd for now.

GPT-6.1 Sol, Ultrafast and Pro 500

OpenAI also announced GPT-6.1 Sol, a cheaper workhorse model it says delivers near-Astra results at a fifth of the price on DeepSWE v1.1. API pricing is $2 per million input tokens and $10 per million output tokens. Cached input costs just $0.10, which should help agents that reuse context. OpenAI claims it beats Opus 5.5 on GDP.pdf and AutomationBench, but those are the company’s own numbers, so treat them with care. Launching a new Sol days after GPT-6 Sol also looks a lot like damage control to me.

Then there is Ultrafast, a new premium speed tier. It offers up to 8x faster generation in Codex (around 300 tokens per second) and up to 6x in the API. GPT-6 Astra Ultrafast is live now on Pro 500 and Enterprise plans, and it costs $60 per million input tokens and $300 per million output tokens. That is not a typo. Leaks pointed to Cerebras hardware, but OpenAI hasn’t confirmed it.

Pro 500 starts at $500 a month and offers 25x the allowance of ChatGPT Plus. It includes Astra Ultrafast, Codex, deep research, image creation, memory and Dots. Dots also work on every Pro tier, including Pro 100. OpenAI has reset all plan allowances too, so everyone gets a fresh slate.

The rest of the list includes Codex in the cloud, computer use in the Agents API, plugin extensions and ChatGPT Space for shared team knowledge. Pages, a new document type for humans and agents, arrives in Q4 2026.

Plenty of launches, then. Whether any of them stops the Anthropic headlines is a question for the next round of benchmarks.

For more updates like this, check out the news section!