Breaking

Sakana AI and 360 launch Mythos-style models amid Anthropic export curbs

June 27, 2026 at 08:02 EDT

Foundation Models
AI Agents
Policy & Regulation

Tokyo-based Sakana AI and Chinese cybersecurity firm 360 rolled out systems billed as "Mythos-equivalent" in late June 2026, as U.S. export controls on Anthropic's most advanced models persisted. On June 22, Sakana released Fugu and Fugu Ultra, a multi-agent platform that dynamically routes across several frontier LLMs, claiming it stands shoulder-to-shoulder with Anthropic's Fable 5 and Mythos Preview (sakana.ai). Around the same time, 360 unveiled Tulongfeng, an automated vulnerability-discovery tool, and Yitianzhen, a cyber-defense automation tool, at the ISC.AI 2026 conference in Beijing (Reuters).

June 27, 2026 · Frontier AI Geopolitics

Asian AI Startups Rush to Fill the Gap as Anthropic's Export Ban Drags On

With US controls blocking foreign access to Anthropic's "Mythos-class" models, China's 360 and Tokyo's Sakana AI launch rival tools — while Zhipu's open-weight GLM-5.2 quietly closes in on closed-source leaders.

JUN 9

Anthropic ships Fable 5 + Mythos 5

First public Mythos-class models.

→

JUN 13

US export-control directive

All models disabled for compliance.

→

JUN 27

Asian rivals launch

Sakana, 360 & Zhipu move in.

GLM-5.2 closes in on closed-source — Terminal-Bench 2.1

Score out of 100. The open-weight challenger lands just 4 points behind the leader.

81.0

GLM-5.2

Zhipu · open-weight

85.0

Claude Opus 4.8

closed-source

744B

MoE parameters in GLM-5.2

token context window

62.1

SWE-bench Pro score

Arena Code (Fable 5 excluded)

The challengers stepping into the gap

Fugu

Sakana AI · Tokyo

Agentic orchestration of multiple model APIs; Japanese optimization. Pitched to Japanese firms & government as "frontier capability without export-control risk."

Tulongfeng / Yitianzhen

360 · China

Vulnerability discovery and automated cyber defense. Framed as a "strategic national asset." Specs undisclosed.

GLM-5.2

Zhipu AI · China

744B MoE, 1M-token context, strong coding/agentic. Open-weight under MIT license.

Optimistic reception

"First open-source model usable as a daily driver."
"Well-timed as a Fable / Mythos alternative."
Strong in long-context and agentic tasks.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.