J.P. Morgan Rating of SmartScale GLM-5.2: Ranks Among Top Global Leaders but Still Faces Challenges to Replace Anthropic

**Zhiku’s GLM-5.2 Secures Global Top 3 LLM Spot, But US Anthropic Restrictions Unlikely to Boost Its Revenue: JFR Report** On June 22, JFR Express reported that Chinese AI firm Zhiku announced its GLM-5.2 large language model (LLM) has entered the global top three for large models, a landmark achievement for China’s AI development—though U.S. restrictions on rival Anthropic’s top models won’t likely drive major revenue growth for Zhiku. In a June 17 Chinese tech industry report from JFR, GLM-5.2 ranks third on the Model Intelligence Index, trailing only Anthropic and OpenAI. This marks the first time a Chinese model has cracked the global top three. The report notes GLM-5.2 excels at programming and long AI workflows, cementing its position as one of China’s highest-performing large models. Zhiku launched GLM-5.2 on June 13 via its Coding Plan, with its API opening on June 17. The model uses a 744-billion-parameter mixture-of-experts (MoE) architecture, with ~400 billion active parameters. It supports a 1-million-token context window and is open-sourced under the MIT license. Per JFR, GLM-5.2 ranks third globally (first in China) on the Model Intelligence Index, fourth globally (first in China) in coding metrics, second globally (first in China) in agentic tools, and its Max variant places second on the Arena.ai Frontend Programming Leaderboard—surpassing Claude Opus 4.7 and 4.8. The report also flags that U.S. restrictions on Anthropic’s Claude Mythos 5 and Fable 5 models may not be sustainable long-term. After Anthropic released these models on June 9, the U.S. imposed access limits citing national security concerns, including rules barring non-U.S. personnel working inside the U.S. from accessing them. Anthropic temporarily suspended global access and tightened identity verification in the short term due to challenges distinguishing user identities. JFR argues that implementing a “U.S.-only access rule for cutting-edge models” faces high execution hurdles and could harm the U.S. AI ecosystem, given many U.S. AI researchers and engineers are foreign-born or non-U.S. citizens. Excluding them from advanced models could slow research progress and push talent and demand overseas. Even so, restrictions on Fable 5 won’t likely translate into meaningful revenue for Chinese AI firms. JFR’s channel research shows most users won’t switch directly from Claude to GLM-5.2, as GLM-5.2’s coding capabilities are comparable to Claude Opus 4.7 but still lag behind Opus 4.8. Many developers adopting Chinese open-source models use them via local deployment, cloud providers, or platforms like OpenRouter instead of directing revenue to model developers. Additionally, developers typically mix models for different tasks (code generation, debugging, long-context work) rather than fully migrating to one, limiting GLM-5.2’s commercial upside from potential Anthropic user overflow. J.P. Morgan adds that Intellispectrum faces a shortage of high-end inference compute power—especially as long-context and agentic workflows demand more resources, hampering its ability to meet enterprise demand. It also notes China’s LLM market is intensely price-competitive: GLM-5.2 delivers a ~26% performance improvement over GLM-5.1, but its API pricing remains unchanged, indicating limited pricing power for model vendors. Long-term risks from U.S. tech controls persist: J.P. Morgan says restrictions on U.S. model distillation technology and exports could limit China’s ability to close the performance gap with U.S. models, widening the divide between Chinese and American LLMs. The core conclusion: GLM-5.2 proves Chinese models have reached the global forefront, but they haven’t yet “filled Anthropic’s shoes.” Constraints including low revenue conversion, compute shortages, price competition, and U.S. tech limits mean Intellispectrum will struggle to gain significant commercial benefits from the Anthropic access restrictions in the short term.