Z.ai GLM-5.2
Z.ai positions GLM-5.2 as its new flagship built for long-horizon agentic coding, a substantial jump over GLM-5.1 (and GLM-5) on multi-step engineering tasks. It is a ~744B-parameter Mixture-of-Experts model (~40B active per token) with a 1M-token context and a new 'IndexShare' mechanism Z.ai says cuts per-token FLOPs by roughly 2.9x at 1M context. Z.ai's own benchmarks show large gains over GLM-5.1 (e.g. FrontierSWE ~74% vs ~31%, Terminal-Bench 2.1 81.0 vs 63.5, SWE-bench Pro 62.1 vs 58.4) and claim it edges out GPT-5.5 on several long-horizon coding suites while trailing Opus 4.8. Weights are released open under the MIT license.
Launched June 13, 2026 to GLM Coding Plan subscribers, with standalone API/chatbot access and open weights (Hugging Face + ModelScope) following within days (~June 16-17). Z.ai markets it as a low-cost open alternative, with reporting citing roughly one-sixth the cost of GPT-5.5 on comparable coding work.
Reality review pending — we publish the executive verdict after the launch has had time to meet the real world.
An open-weights Chinese flagship that credibly contests top US closed models on agentic coding at a fraction of the price signals that the open/proprietary and US/China capability gaps are narrowing fast. For executives, it reframes build-vs-buy and vendor lock-in calculus for coding and agent workloads.