In my February 18 note, we called the race for AI supremacy a two-horse race, winner-take-all, and “very, very tight.”

Four months later, that statement still holds.

Today Anthropic released Fable 5 to the public. This is the guardrailed version of Mythos. 

Remember, Mythos is the Anthropic model we discussed in April that found and exploited decades-old security flaws on its own. It was deemed so capable they didn’t release it, and instead took it to the government and convened industry leaders to start building protection against it.

That said, around 1pm this afternoon, they announced the release of Fable 5 as the “safe” version of Mythos.

Stocks bottomed right around that time, after what was becoming another messy unwinding day for AI stocks.

So now we have another big leap forward in model capabilities — maybe the most consequential since the early February models that brought the explosion in agentic AI.

With that, the read over the past half year or so has been that — Anthropic is king, AGI is near (if not here), the U.S. has won the AI race. 

But this chart says otherwise. 

This tracks the “Intelligence Index” of the world’s top AI models (the dots).  Notice, the U.S. models are on top. But what’s clear, is that the dots are clustering tighter and tighter as the model intelligence is rising.

The frontier isn’t separating, it’s converging. OpenAI, Anthropic, Google, xAI are bunched together at the top. And the Chinese labs –DeepSeek, Alibaba, Kimi have compressed the gap, not widened it.

What’s the finish line? 

Human-level intelligence, that becomes self-improving.

The first to that finish line should be the winner, and therefore, should be in position to set standards, attract talent, and determine what technology gets embedded into governments and critical infrastructure.

And as we’ve said before, that will mean the difference between AI that serves humanity, or AI that controls humanity (serving the interest of the Chinese Communist Party).

Is Mythos (and today’s publicly available model, Fable) the model that reaches the finish line?

Maybe. Consider this: Anthropic preceded today’s model release with a note last Thursday titled, When AI builds itself (see it here).

And Anthropic’s founder Dario Amodei had already described Mythos a few months ago as “by far the most powerful AI we’ve ever built.” Around that time, he also did a podcast where he said, the surprising thing isn’t how fast they’re accelerating, it’s “how close we are to the end of the exponential.”

Not the start. The end. He’s saying the steep part of the climb is nearly behind us.

If he’s right, the convergence on that chart above may not be a pause before someone breaks away. It may be the field arriving near the finish line together.

So, still a tight race.

Which is why China is a significant threat. One Chinese model leap doesn’t have far to travel.