Fable 5 and Mythos 5 Stalled: Asia Unveils Its AI Alternatives

Fable 5 and Mythos 5 Blocked: Asia Unveils Its AI Alternatives

Two weeks after the global suspension of Fable 5 and Mythos 5, the two most advanced models from Anthropic, two Asian companies have launched alternatives. In Japan, Sakana AI has released Fugu Ultra, a multi-agent system focused on orchestration. In China, cybersecurity firm 360 has unveiled two tools for vulnerability detection and incident response.

Mythos 5 is unusable by any foreign national, wherever they are in the world.

What were you doing on June 12? Across the Atlantic, the Bureau of Industry and Security, part of the U.S. Department of Commerce, sent Anthropic an export control directive suspending access to Fable 5, its consumer model, and Mythos 5, its experimental version with restricted access, for any foreign national, wherever they are in the world. Anthropic received the letter at 5:21 PM New York time, without further explanation regarding the national security reasons cited. The California lab complied with the directive, albeit contesting it. They argued that the discovery of a partial and non-universal jailbreak did not justify recalling a model deployed to hundreds of millions of users. Clearly, competitors took advantage of Anthropic's sidelining to emerge.

In the days that followed, two companies launched substitute products.

Sakana AI and Fugu Ultra

On June 22, Sakana AI launched Fugu, in two versions: Fugu for everyday use, and Fugu Ultra for complex tasks. The Tokyo-based startup was founded in 2023 by Ren Ito and Llion Jones, two former researchers from Google, who were later joined by David Ha, also a former Google employee. The name refers to the fugu, the Japanese pufferfish. On major engineering, reasoning, and science benchmarks, Fugu Ultra demonstrated performance comparable to Fable 5 and Mythos Preview, according to data published by Sakana AI.

The user sends a request to a single endpoint, and the model decides whether to handle the request directly or to mobilize several specialized agents in parallel. Fugu is a language model trained to coordinate other models, without a fixed pipeline. Trinity and Conductor, two research studies presented at ICLR in the spring, served as the foundation for this system.

On its website, Sakana AI showcased "cutting-edge capabilities without the risk of export controls." David Ha wrote on X that access to the best models can disappear overnight, and that reliance on a single provider for national infrastructure is a risk that recent export controls have made impossible to ignore. According to a spokesperson for Sakana AI, the launch was "entirely coincidental" with Anthropic's suspension, as work on Fugu had started the previous year.

360 and Its Cybersecurity Tools

Chinese firm 360 has launched two AI tools against software vulnerabilities and cyberattacks. 360, a Beijing-based cybersecurity company, presented two tools. The first, Tulongfeng, is dedicated to the automatic discovery of software vulnerabilities. The second, Yitianzhen, covers defense and incident response. Zhou Hongyi, founder of 360, stated during the presentation that these capabilities constitute a national strategic asset, and warned against what he calls "one-way transparency," where some have access to advanced vulnerability detection tools while others are excluded.

360 has not published any benchmarks comparable to those of Sakana AI, and did not respond to requests for comments from TechCrunch.

Anthropic had reported in May an annualized revenue of over €42 billion. No one knows what share of that comes from Asian enterprise clients. At the G7 summit in Évian the previous week, Ren Ito expressed hope that the United States would preserve access to cutting-edge models for its closest allies.