The first model of Ilya Sutskever is probably really coming.
Yesterday, a 16z partner, Martin Casado, made a sudden announcement: "I just got access to a new model. This is going to be the most important model release of the year, and it's even possible to get rid of one".
Casado did not disclose the model name, but within a few hours, more news began to emerge.
Scattered clues point to one answer: Ilya Sutskever's new model.
The first to announce this was Martin Casados, a 16z partner in charge of infrastructure investments. A 16z, as one of the capitalists of this mysterious lab, logically contacted the new model of SSI in advance.
Later that day, Andrew Curran, co-founder of the AI news website The Rundown, brought a related message.
He said he knew little about the model and had only unverified rumours. But if he focused on the AI industry in the long run, he could tell which messages "sound like true".
But he pointed out that a non-head laboratory has made a breakthrough in continuous learning.
As early as early August, Gavin Baker, founder of Atreides Management, revealed on the podcast Invest Like the Best that SSI plans to release their first model in August.
Two years after leaving OpenAI, Ilya's company is finally quiet.
In May 2024, Ilya left OpenAI after nearly a decade of work.
A month later, he co-founded SSI with Daniel Gross, Daniel Levy. The company's name is directly derived from its ultimate goal: Safe Superintelligence, security superintelligence.
Since its inception, SSI has distanced itself from other model companies.
The whole company has only one goal, one product and one roadmap. It does not pursue frequent product updates or rush to update the model list before competitors, but instead advances capabilities and safety along the same technological path.
SSI also gives this model a very Ilya-style saying: scale in peace in silence".
Over the next two years, SSI virtually disappeared from the public eye. There were no public research results, and the official website had only a brief mission statement and recruitment entries for the long term.
However, capital markets are willing to pay a staggering price for this silence.
In 2025, SSI completed $2 billion in financing and was valued at $32 billion. It has no revenue and no products, but has become one of the highest-valued AI startups in the world.
SSI is really open to "product description" and for a long time only interviewed Ilya himself a few times.
In a lengthy conversation with Dwarkesh Patel in November 2025, he summarized 2012 to 2020 as the "Research Era" and 2020 to 2025 as the "Age of Scale". Continuing to add data, parameters and algorithms will still bring progress, but it is difficult to replicate the fundamental leaps of the past decade.
Today, AI is returning to the age of research.
Ilya argues that one of the most fundamental problems with current models is their ability to generalize much less than humans. They can be trained in bulk data and can outperform the vast majority of people on some tasks, but it is difficult for them, like humans, to quickly master new capabilities from a few experiences, feedback and demonstrations.
He also used a 15-year-old to explain the super-intelligence he wanted.
A highly intelligent 15-year-old will not be born a doctor or a programmer, but will gradually master these skills through learning and practice as soon as he enters the workplace. The same is true of Ilya's super-intelligence: it doesn't have to know everything at birth, but it needs the ability to continue learning after entering the real world.
After two years of low ratings, the company has finally stepped up significantly this year.
In July, Invisalign announced a long-term partnership with SSI to acquire Invisalign's latest Vera Rubin hardware. Reuters reported that Invisalign would also invest about $5 billion in SSI.
Ilya seldom revealed at the time that SSI had "deserved to scale" the results of their research, and that a large enough British computer would help them do so.
"Continuous learning" is still the biggest suspense.
Currently, there is very little public information about the mysterious model's capabilities. Casado has not revealed any test results, and the continuing learning breakthroughs cited by Andrew Curran have not been temporarily verified.
Continuing learning has become the focus of this round of speculation, largely because it is highly aligned with the direction of research that Ilya has previously publicly discussed.
This is the core capability of the 15-year-old in Ilya's imagination: the starting point does not have to cover all knowledge, but must be able to continue to learn through deployment and work.
It's not easy to really achieve it.
When models learn new knowledge, they can destroy the original parameters; learn a new task and forget the ability to master it. This phenomenon is known as "catastrophic forgetting".
Updates are too cautious, models cannot be learned effectively; updates are too radical, original capabilities, behavior patterns, and even security boundaries can change unpredictably.
If a laboratory really solves continuous learning, big models can change the way they are trained, deployed, and updated to date.
This may have been Casado's "Most Important Release of the Year".
I'm looking forward to SSI revealing the answer personally.
The link is as follows: