Gemini 4 Argon tops the Vals Index, but the real benchmark is the customer

By Nate from Nate's Substack · October 7, 2026.

Google has unveiled Gemini 4 Argon, a new model with benchmark results the company itself calls impressive. According to Google, Argon leads the Vals Index, an index that evaluates performance on professional work across several fields. For now, the initial rollout is limited: it is only available to trusted "cybersecurity defenders", with plans to widen access later. Google has also shared descriptions of internal uses and of early work with some customers, but, as the author notes, most users still cannot try it for themselves.

Nate says he wants Google to succeed and wants powerful AI in more people's hands, and declares himself very optimistic about today's builders: founders, people building inside existing companies, and those who know one corner of the world so well that they can spot where everyone else is still struggling.

The question the author wants to answer is not so much what score Argon gets on a benchmark, but what it will actually let a specific customer do: where they will come across it, which tasks will become easier thanks to it, and what that customer will be willing to delegate to it next time. In an earlier video (mentioned but not detailed in this email), Nate had already argued that customer obsession is the real benchmark that matters, above model leaderboards. In this piece he says he takes that idea further: what exactly paying attention to customers lets you notice, and how that observation can be used to find something worth building.

The post (paid, as the email itself indicates) promises to develop four concrete ideas:

- Pay attention to what people want to delegate: every failure in a delegated task reveals the exact point where the customer still has to do the work themselves.

- Look for the work people have already abandoned or given up on: the questions that bring to light what a customer stopped expecting to get.

- Let the customer's real life shape the product: the author mentions, by way of example, what a walk in the woods teaches him about building products that go beyond the classic text box.

- Three questions to apply to whatever you are building yourself, aimed at checking whether you really understand someone else's problem and what you have changed in your product after watching them try to use it.

The email closes with a central idea from the author: useful intelligence is already available today and we have barely begun to turn it into well-thought-out products. The body of the message is, in essence, a teaser for the full article, which is reserved for paying subscribers, along with access to "Nate's Library MCP" and his Slack community. The email includes no further technical details about Gemini 4 Argon, no specific Vals Index figures, and no additional examples of the customer use cases mentioned by Google.