CEO of Chinese Anthropic rival tells Elon Musk that China will have a Fable 5-class AI model before next year — it ‘won’t take that long’ says Jie Tang in response to Musk's prediction of a Q1 target

Elon Musk SpaceX
(Image credit: Getty / Spencer Platt)

The CEO and founder of one of China's leading AI startups has warned that a rival Chinese Fable 5-class AI model is closer than even Elon Musk thinks, in a recent X interaction. In a post on his own platform, Musk stated that China's attempt at an AI model to rival Anthropic's new Fable 5 offering will arrive in "Probably Q1," to which Jie Tang replied, "won't take that long."

Jie Tang might be a relative unknown in the Western world, but he’s the founder of Z.ai, formerly known as Zhipu AI, and is one of the leading AI startups in China, based in Beijing. It released its latest AI model, GLM-5.2, on June 16, 2026, and the company’s benchmarks show that it has almost the same performance as Anthropic’s Opus 4.7 to 4.8, which launched in April and May of this year, respectively. It has also consistently outperformed OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro.

Anthropic launched its latest, most powerful public model, Fable 5, last June 10, which itself is a nerfed version of the Mythos 5 that was previewed to select entities in early April. However, three days after the general public could start enjoying Fable 5, the U.S. government put an export control directive on the model and barred all foreign nationals, including Anthropic’s own employees, from accessing it. Since the company determined that it cannot guarantee total compliance with the order, it decided to pull both Fable 5 and Mythos 5 from across the globe.

Latest Videos From

If Z.ai is indeed working on a more powerful frontier AI model that could compare against or even outperform Fable 5, users who want the latest, cutting-edge models might just switch over to the Chinese AI provider once it goes online. This does not mean that the U.S. would automatically be disadvantaged, though, as it still has Mythos and Fable. The U.S. government claimed that it only put the export control Fable 5 after Amazon allegedly discovered that the Anthropic model’s guardrails could be broken, and that the company refused to address it before going live. In its defense, the AI startup claims that the so-called jailbreak is minor and is replicable in other models like GPT-5.5. Washington claims that once the issue has been resolved through a patch, it will lift the restrictions and allow foreign users to access Fable 5 again.

Both China and the U.S. are in a race to get the most advanced AI model possible. The U.S. has put major roadblocks to block China’s progress, including the imposition of export controls on the latest AI chips, tools needed to make the latest semiconductors, and even the software needed to design them. Despite that, Chinese firms find ways to achieve breakthroughs, with one of the biggest developments so far being the arrival of DeepSeek in late 2024. It seems that Jie Tang wants to take the Chinese crown away from DeepSeek by claiming that Z.ai could come out with a model that could match the United States’ best in the coming months.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Jowi Morales
Contributing Writer

Jowi Morales is a tech enthusiast with years of experience working in the industry. He’s been writing with several tech publications since 2021, where he’s been interested in tech hardware and consumer electronics.

  • S58_is_the_goat
    You can't have nuclear weapons or ai better than America, time to invade! Hooooooooo...
    Reply
  • Arkitekt78
    And it will be just as crappy as everything else China makes.
    Reply
  • phead128
    GLM 5.2 already beats Fable according to this user preference coding ranking, and at a fraction of cost + open weight.

    https://medium.com/data-science-in-your-pocket/glm-5-2-beats-claude-fable-5-glm-5-2-benchmarks-explained-493751c8a24f
    A higher Elo indicates users generally prefer the generated coding solutions.

    That's the problem is "banning" Fable when Chinese open source is highly competitive at much cheaper price point. If you are doing agentic work, you can use the Chinese models for bulk of work and use frontier Western models for planning or stubborn debugs given the high token costs.
    Reply
  • circadia
    Arkitekt78 said:
    And it will be just as crappy as everything else China makes.
    I'd be more cautious if I were you. China's not going to take over the world sooner or later, but it also isn't going to stay forever just minnows, is it?
    Reply
  • ivan_vy
    Arkitekt78 said:
    And it will be just as crappy as everything else China makes.
    the key is availability, Kimi 2.6 is equivalent to GPT 5.4 but is "free" (harvest your data) and the context window is huge. for menial tasks is enough.
    Reply
  • LordVile
    And full self driving was in what year?
    Reply
  • BloodLust2222
    Funny, I bet you didn't know that 80% of what you use in daily life comes from China. Just because the stuff you buy off Temu is junk, Doesn't mean everything made in China is junk. I've had better quality stuff from their than some garbage Americans put out.
    Reply
  • phead128
    ivan_vy said:
    the key is availability, Kimi 2.6 is equivalent to GPT 5.4 but is "free" (harvest your data) and the context window is huge. for menial tasks is enough.
    Kimi 2.6 on OpenRouter API does not train on your data and its also open source so you can self host your own solution or local infrastructure, which is definitely something Claude or OpenAI can't offer. You can even fine tune Kimi 2.6 to be domain specific which Claude/OpenAI can't offer.
    Reply
  • Pierce2623
    Of course they’ll just “distill” it. Of course the definition of “distillation” is quite open when Deepseek contained so much GPT3.5 code that if you asked deepseek R1 to compare itself to ChatGPT3.5 it would say “I don’t understand. You’re asking me to compare myself to ChatGPT3.5 but i am ChatGPT3.5” lol. Yeah “distilled” huh? Thats a nice euphemism for “stole the code”.
    Reply
  • phead128
    Pierce2623 said:
    Of course they’ll just “distill” it. Of course the definition of “distillation” is quite open when Deepseek contained so much GPT3.5 code that if you asked deepseek R1 to compare itself to ChatGPT3.5 it would say “I don’t understand. You’re asking me to compare myself to ChatGPT3.5 but i am ChatGPT3.5” lol. Yeah “distilled” huh? Thats a nice euphemism for “stole the code”.
    How can it be "distilled" when Fable is banned for all users?

    The whole copy/steal narrative only works when the model is available, it's currently not available.
    Reply