GLM-5.2

Zhipu / Z.ai's flagship coding-first agentic model and successor to GLM-5.1 — a ~753B MoE with DeepSeek Sparse Attention (DSA) and the highest-scoring open model on the AA Intelligence Index v4.1 (51, #1 of 92). Launched 2026-06-13 on the GLM Coding Plan tiers; full MIT open weights, standalone API, and benchmarks followed 2026-06-16. First GLM with a "solid" 1M-token context (131,072 max output), with High/Max thinking-effort presets (Max recommended for coding).

Architecture: introduces IndexShare (arXiv 2603.12201) — reusing one lightweight sparse-attention indexer across every four layers to cut per-token FLOPs 2.9× at 1M context — plus an improved MTP layer that lifts speculative-decoding acceptance length up to 20%. Self-reported benchmarks: AIME 2026 99.2, GPQA-Diamond 91.2, IMOAnswerBench 91.0, SWE-bench Pro 62.1, Terminal-Bench 2.1 82.7 (best harness), DeepSWE 46.2, FrontierSWE 74.4 — a large jump over GLM-5.1 and competitive with Claude Opus 4.8 / GPT-5.5 / Gemini 3.1 Pro at far lower cost. Pure MIT license, no regional limits.