Zhipu AI’s open-weights model posted near-frontier coding, agentic and some cybersecurity results, with researchers saying it matched Anthropic’s Mythos in certain bug-finding tests while undercutting leading U.S. models on price.
Z.ai, formerly Zhipu AI, has released GLM 5.2, an open-weights model that CNBC and security researchers said is approaching top U.S. systems on coding, agentic and some cybersecurity tasks while costing far less to use. The Beijing-based model scored close to Anthropic’s Opus 4.8 on key benchmarks, outperformed it on MCP-Atlas, and researchers said it matched Anthropic’s Mythos in some security bug-finding scenarios, though it still trails Anthropic and OpenAI on other tasks. GLM 5.2 launched to GLM Coding Plan subscribers on June 13, with open weights and benchmark scorecards released under the MIT license on June 16. The 744-billion parameter mixture-of-experts model has a 1 million token context window, runs at 300 tokens per second, and is priced at $1.40 for input and $4.40 for output per million tokens, versus $5 input and $25 output for Anthropic’s Opus 4.8. The timing drew attention because Anthropic disabled Fable 5 and Mythos 5 globally after a June 12 U.S. Commerce Department export-control order. Analysts and researchers said the release highlights intensifying AI competition and could add pressure as the White House reworks U.S. AI policy, while security and data-governance concerns remain around PRC-origin models.