zai-org/GLM-5.2

GLM-5.2: Long-Horizon Task Specialist

zai-org's GLM-5.2 is their latest flagship model, significantly advancing long-horizon task capabilities, particularly with its solid 1M-token context window. This marks a substantial improvement over its predecessor, GLM-5.1, ensuring stable performance for extended tasks.

Key Capabilities

• 1M-Token Context: Reliably handles and sustains long-horizon workloads with an extensive context window.
• Advanced Coding: Offers enhanced coding capabilities, allowing for flexible effort levels to balance performance and latency.
• Improved Architecture: Features IndexShare, which reuses the same indexer across every four sparse attention layers, leading to a 2.9x reduction in per-token FLOPs at 1M context length. It also includes an improved MTP layer, boosting speculative decoding acceptance length by up to 20%.
• Pure Open-Source: Released under an MIT license, ensuring unrestricted technical access globally.

Performance Highlights

GLM-5.2 demonstrates strong performance across various benchmarks, particularly in coding and reasoning tasks. It shows notable gains over GLM-5.1 and competes with other leading models in categories like SWE-bench Pro, NL2Repo, and Terminal Bench 2.1. For instance, it achieves 62.1 on SWE-bench Pro and 81.0 on Terminal Bench 2.1 (Terminus-2).

Good For

• Applications requiring extensive context understanding and generation.
• Complex coding tasks where balancing performance and latency is crucial.
• Developers seeking an open-source model with strong long-horizon capabilities and architectural innovations.