Zhipu AI open-sourced GLM-5.2 under MIT license on June 15, 2026. The model features a 1 million token context window, directly countering US export restrictions on Anthropic's Claude models.
Key facts
- GLM-5.2 released June 15, 2026 under MIT license.
- 1 million token context window.
- Direct response to US export restrictions on Anthropic.
- Zhipu AI raised $400M at $3B valuation in May 2024.
- Largest open-weight context window among competitors.
Zhipu AI announced GLM-5.2 will be released under MIT license with 1M token context as a direct response to US export restrictions targeting Anthropic models According to Pandaily. The Beijing-based AI lab, which raised $400 million in May 2024 at a $3 billion valuation, is positioning the open-weight release as a geopolitical hedge: any developer worldwide can now run a frontier-class model locally without US export controls.
The 1M Context Window
GLM-5.2's 1 million token context window matches Anthropic's Claude Opus 4.7, which launched in May 2026. The MIT license permits unrestricted commercial use, modification, and redistribution, a stark contrast to Anthropic's proprietary licensing. Zhipu AI did not disclose training compute or benchmark scores, but the company claims the model achieves "competitive performance" on Chinese-language reasoning tasks. The open-source release includes model weights, inference code, and a fine-tuning framework.
Geopolitical Context
The move follows the US government's June 13 order for Anthropic to shut down its strongest Claude models, citing national security concerns [per gentic.news reporting]. By open-sourcing GLM-5.2, Zhipu AI ensures Chinese developers and global partners retain access to high-capability AI without relying on US-licensed models. The timing also coincides with Nvidia denying Anthropic's claims of chip smuggling via Latin America [gentic.news, June 13].
Competitive Landscape
GLM-5.2 enters a crowded open-source field. Meta's Llama 4.5, released in April 2026, supports 512K tokens under a custom commercial license. Alibaba's Qwen3.5 offers 256K tokens under Apache 2.0. Zhipu AI's 1M token window is the largest among open-weight models, though Google's Gemini 2.0 Ultra (proprietary) supports 2M tokens. The MIT license is more permissive than Llama's, potentially accelerating adoption in enterprise and research settings.
What to watch
Watch for benchmark results on Chinese-language reasoning tasks (C-Eval, MMLU-CN) and whether GLM-5.2 adoption in enterprise applications exceeds 100K downloads within 90 days. Also monitor US export control adjustments in response to open-source workarounds.
Source: pandaily.com
