Anthropic’s Fable shutdown is a big moment for open-source AI

Anthropic’s suspension of its top AI models late last week revealed a harsh reality for the companies that rely on them: access can be cut off at any time.

It’s been a big topic this week, and one that Wall Street will be watching closely as Anthropic and OpenAI prepare for potentially massive IPOs in the coming months.

Microsoft CEO Satya Nadella warns of risks even though his company is OpenAI’s primary investor and backed Anthropic with billions of dollars last year. wrote on monday to post In X, companies must “create intermediary systems that evolve over time while maintaining control over their IP.”

“The last thing any of us wants is a world where every company in every industry values ​​a few models who eat everything in sight,” Nadella wrote.

Investors are trading on this theme. MiniMax And ZhipuChallenge Anthropic, China’s open-source AI lab, took off on Monday as it shed light on downloadable models that companies can run themselves.

The Anthropic-Fable news came at an odd moment late Friday. About two hours ago, SpaceX it surrounded him first day of trading After the largest IPO in history, SpaceX’s xAI unit is a niche player in the artificial intelligence space, but CEO Elon Musk is an outspoken voice on the subject.

anthropic announced It had removed access to the Fable 5 and Mythos 5 models to comply with the US government’s export control directive, which cited “national security authorities”. Anthropic abruptly disabled the models for all its customers to ensure compatibility, but said none of the other models would be affected.

There is another approach for developers who want full control over model access. An open source model can be downloaded, run on the company’s own infrastructure, and customized based on the company’s data and needs. When the model lives on a company’s own servers, no political struggle can shut it down.

Yash Patel, CEO of Applied Compute, which helps companies train and run custom models, said the Anthropic challenge “highlights the importance of having your own model.” He said change has become much more common lately.

“The fact that they want a multimodal future is probably what we’ve been hearing more and more in the last month, all year long,” Patel said. “They don’t want to be tied to one vendor.”

This could be a problem for the US, as the most adopted open models come from China, just as the world’s two largest economies are struggling to control the future of AI.

Models from DeepSeek, TencentXiaomi and MiniMax are among OpenRouter’s most used this month, despite their closed competitors. Zhipu framed his latest statement as a rebuttal of sorts to Washington. The company argued that cutting-edge AI should not be owned by a handful of players or withdrawn at will.

Cost can also accelerate adoption. As the price of cutting-edge AI rises, companies are shifting routine tasks to cheaper models and saving the most expensive ones for the most difficult tasks.

Patel said customers are reacting to what he calls a “token apocalypse” as AI products move toward usage-based pricing.

“The era of token maximization is over,” he said. Companies are now looking for “better, cheaper, faster models.”

This is prompting some businesses to rethink models they would have rejected months ago, including open models from China.

“Before, I was like I didn’t even want to talk about it,” Patel said. “Now they’re saying: OK, how good can it be, and if it’s good, we’ll figure it out.”

This is a reminder that the AI ​​market is still in its infancy, with ChatGPT going public less than four years ago. For investors, this reframes who is actually leading the AI ​​race. Despite what current valuations suggest, it may not be just the big closed model labs that are the winners.

WRISTWATCH: DOD’s Emil Michael on Anthropic action