Zhipu’s GML 5.2 Matches Frontier AI Models as Open-Source Pressure Builds Before 2 IPOs
Updated
Updated · Financial Times · Jun 25
Zhipu’s GML 5.2 Matches Frontier AI Models as Open-Source Pressure Builds Before 2 IPOs
3 articles · Updated · Financial Times · Jun 25
Summary
GML 5.2 became the first open-source model this month to reach parity with leading frontier AI systems on widely watched benchmarks, and it has also drawn strong reviews as a coding agent.
That performance threatens a key edge for Anthropic and OpenAI by suggesting their software harnesses and premium model packaging may be less durable differentiators than investors assumed.
Customer sticker shock over rising AI bills is already pushing less demanding workloads toward cheaper models, while intermediaries such as Japan’s Fugu Ultra promise to split tasks across multiple suppliers.
Anthropic’s shutdown of Fable 5 access under pressure from the Trump administration has further encouraged buyers to avoid dependence on one provider and boosted the appeal of open-source models that can run on private servers.
The shift raises the stakes for Anthropic and OpenAI ahead of IPOs, as they must defend pricing power, curb model copying and satisfy Washington while open-source rivals close the gap.
Are U.S. restrictions on its own AI creating an unstoppable Chinese competitor?
As AI models learn to steal from rivals, can new digital watermarks prevent IP theft?
2026 AI Power Shift: US Export Bans on Anthropic Spark Global Race for Sovereign and Open-Source AI
Overview
In June 2026, the US government imposed strict export controls on Anthropic's advanced AI models, Claude Fable 5 and Mythos 5, citing national security concerns about their potential misuse and jailbreaking. This action came just after Anthropic’s CEO publicly supported government authority to block unsafe AI models. Many at Anthropic saw the move as a response to past conflicts with the Department of Defense, especially since Fable 5’s safety guardrails were already considered overly strict. The unusually broad order disrupted global access to Anthropic’s AI, exposing the risks of relying on US-based AI and sparking calls for greater AI independence worldwide.