OpenAI CFO Sarah Friar laid out the company’s growth figures and its approach to releasing models on the day of DevDay. Overall revenue grew more than 70% quarter over quarter in Q3, she said, and business-to-business revenue has doubled since the start of July. The newly announced dots agents will come first to ChatGPT Pro and enterprise offerings, while GPT-6.1 Astra, the next frontier model, is being held back on safety and alignment grounds, meaning the work of keeping a model’s behavior in line with intended goals.
The growth figures Friar shared
| Measure | Friar’s account |
|---|---|
| Overall revenue | More than 70% growth quarter over quarter in Q3 |
| Business-to-business revenue | Doubled since the start of July |
| Consumer reach | 1.2 billion weekly active users |
| Enterprise reach | 2.5 million organizations using OpenAI platforms |
| Capital raised | $122 billion in Q1 |
These figures use different baselines and time frames. The revenue growth rate compares one quarter with the previous one, while the business revenue figure compares the present with the start of July. All of them are the company’s own figures.
Friar attributed consumer growth partly to improved image and voice features and to seasonal back-to-school and back-to-work use. On the enterprise side, she pointed to dedicated sales teams and offerings tailored to coding, cybersecurity and life sciences. For small and medium-sized businesses, OpenAI introduced a $100-per-month tier designed for all-day workflow use instead of five-hour usage intervals.
OpenAI’s dots connect to the tools people already use
Friar described AI as moving past answering questions and reasoning toward proactive help with daily work. OpenAI’s dots are at the center of that shift. An AI agent can take on a continuing task and act on information from connected tools, and dots draw on the email, calendars and Slack channels a user connects to surface work that might otherwise slip. One example is a reminder to finish a drafted LinkedIn post scheduled for the next day. They can also use context from past ChatGPT conversations and a user’s habits.
▲ Proactive assistance across work tools
OpenAI plans to introduce dots through ChatGPT Pro and enterprise offerings before expanding to its broader base of 1.2 billion users. Against competing products such as Meta’s Muse, Friar said OpenAI’s edge lies in its underlying models, with GPT-6.1 Sol offering top-tier intelligence at one-fifth the cost of Astra. She also cited distribution across 1.2 billion consumers and 2.5 million organizations as a lasting advantage. Those are the company’s stated strengths, not a guarantee that every reminder or workflow will be useful.
Pacing model releases for safety
Friar said safety and alignment have always been central to OpenAI’s development. In line with that, OpenAI continues to advance GPT-6.1 Sol and GPT-6.1 Luna while deliberately holding back GPT-6.1 Astra, its next frontier model, meaning one of its most capable. The explanation is not a release date for GPT-6.1 Astra, and it is not proof that any model is free of risk.
Enterprise concerns are concrete. Companies worry that autonomous agents could cause unexpected failures inside their systems, and Friar named cybersecurity as the leading concern among corporate leaders. With open-source models advancing quickly, she argued that the window for defenders is closing. OpenAI is working with executives and partners to use models to review corporate infrastructure, find vulnerabilities and build long-term security playbooks.
▲ Paced model releases and enterprise defense
Compute and custom chips
Friar called computing capacity both a competitive advantage and a hard requirement: without compute, there is no revenue. OpenAI works with multiple infrastructure partners to secure reliable capacity and negotiate terms, and it is developing its own hardware under a first-party chip program to improve the efficiency of inference, the process of running a model to produce a response. On higher interest rates, she said they increase the yield on OpenAI’s cash reserves, pointing to the $122 billion raised in Q1.
What organizations should check
Before adopting a proactive agent such as dots, decide which work tools to connect and what access each one should have. With cybersecurity named as a top concern, model-assisted vulnerability reviews belong inside an established security process, coordinated with the partners responsible for it. Smaller businesses can compare the all-day design of the $100 tier with how their own work actually runs. The growth figures and release plans come from the company, so actual terms should be checked in each plan’s own documentation.