Chinese lab Zhipu AI released GLM-5.1 on 7 April 2026 under the MIT licence, claiming it outperforms both Claude Opus 4.6 and GPT-5.4 on SWE-Bench Pro, the expert-level real-world software engineering benchmark. The model has 744 billion parameters in a Mixture-of-Experts architecture with 40 billion active per forward pass and a 200,000-token context window. The MIT licence permits unrestricted commercial deployment, fine-tuning, integration and redistribution. The release creates a direct tension with Anthropic’s decision to withhold Mythos Preview on safety grounds: while the most capable Western frontier model is locked behind a 50-organisation firewall, a Chinese open-source model claiming comparable benchmark performance is freely downloadable globally, including by actors with no commitment to responsible deployment. The development directly affects LE procurement calculus on AI tools and the feasibility of access controls on offensive AI capability.
Partnered with
European Network of Law Enforcement Technology Services
Thursday, April 23, 2026