OpenAI adds task router to Dots to cut compute as Musk calls Grok Bot costs 'unsustainably high'

OpenAI announced today that Dots, its personal assistants, will route tasks to different underlying models, including GPT-6.1 Terra, to reduce how much compute the product consumes. The announcement comes in the same week Elon Musk said Grok Bot is too expensive to run at current revenue, according to a Fortune report.

What OpenAI announced

Terra is smaller and uses less computing power than GPT-6 Astra, the larger model that originally powered all Dots tasks. The new automatic task router will let OpenAI reduce how compute intensive Dots is, Fortune reports. Pricing and benchmark figures for GPT-6.1 Terra were not disclosed.

At a dinner with reporters in New York City, Adam Fry, head of ChatGPT's consumer business, said Dots are "on the compute heavy side." Each Dot has its own cloud computer, which it uses to execute tasks for the user, accessing any accounts and services it has been given. Dots currently help users read and respond to emails and find and book event tickets, among other tasks.

Fry said a "background mode" feature is meant to make Dots an always-available presence. Dots are currently limited to Pro subscribers, at $100 per month. Fry said he expects a breakthrough in compute efficiency that will cause costs to fall rapidly even as model performance improves.

Fry also cited figures for other ChatGPT products:

• ChatGPT's voice mode has 250 million weekly active users.
• OpenAI claims ChatGPT search volume, measured by queries with search intent, is now one-fifth of Google's volume.

Fortune notes that consumer devices designed by Jony Ive are set to debut this year.

Grok Bot costs

Elon Musk said the cost of running Grok Bot, the personal AI assistant SpaceX introduced in August, is "unsustainably high relative to revenue." He suggested the company was throttling back the compute allocation for the technology, causing users to hit their limits faster.

Compute and funding backdrop

Fortune frames Meta's Muse personal assistant as the product that convinced the industry that proactive personal agents are the next step, while questioning whether enough compute exists to power them on the timeline companies are betting on.

An updated amendment to Anthropic's S-1 filing this week lists purchase agreements totalling $895.4 billion, "substantially all of which is related to compute and infrastructure." Inability to secure more compute is listed as a major risk factor. Anthropic's infrastructure includes four different types of chips.

OpenAI has not filed an S-1 or confirmed IPO timing. Fortune's sources say it almost certainly faces similar compute challenges. Its fleet is largely dependent on Nvidia GPUs, and it is developing an in-house chip called Jalapeno, set to scale up production in 2027. According to the Wall Street Journal, OpenAI's latest funding round landed at $11 billion against a $30 billion target.

OpenAI also began referring to data centers as "compute campuses" this week, amid what Fortune describes as nationwide, bipartisan backlash. In its first post using the term, OpenAI promised to reduce strain on local electrical grids and not take limited power supply from local families.

What this means

The Dots router is a direct response to the cost problem Fry described: when every task runs on the largest model, an assistant with its own cloud computer is expensive to serve. Sending tasks to the smaller GPT-6.1 Terra is OpenAI's stated way to cut that load without removing features.

Musk's comments show that cost pressure on personal agents is not limited to one product. In the case of Grok Bot, the stated consequence is that users hit limits faster. For Dots, the stated consequence is access restricted to the $100 per month Pro tier.

Fry's expectation that costs will fall rapidly is a forecast, not a demonstrated result. Fortune also points out the opposing pressure: the more powerful and constantly running an agent becomes, the more compute it uses.

The cost of delivering these agents at scale remains tied to compute supply, which is why the Anthropic S-1 figures and OpenAI's funding shortfall feature in the story.

Source: Fortune.