name = "GLM-5.2"
description = "Open flagship GLM for long-horizon coding agents and million-token context work"
family = "glm"
release_date = "2026-06-13"
last_updated = "2026-06-13"
attachment = false
reasoning = true
temperature = true
tool_call = true
structured_output = true
open_weights = true

[limit]
context = 1_000_000
output = 131_072

[modalities]
input = ["text"]
output = ["text"]

[[weights]]
label = "Hugging Face"
url = "https://huggingface.co/zai-org/GLM-5.2"

[[benchmarks]]
name = "SWE-Bench Pro"
score = 62.1
metric = "resolve rate"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "Terminal-Bench"
score = 82.7
metric = "success rate"
harness = "Claude Code"
version = "2.1"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "FrontierSWE"
score = 74.4
metric = "dominance"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 40.5
metric = "accuracy"
dataset = "text-only subset"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "Humanity's Last Exam"
score = 54.7
metric = "accuracy"
variant = "with tools"
dataset = "text-only subset"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "CritPt"
score = 20.9
metric = "accuracy"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "AIME"
score = 99.2
metric = "accuracy"
version = "2026"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "HMMT"
score = 94.4
metric = "accuracy"
version = "November 2025"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "HMMT"
score = 92.5
metric = "accuracy"
version = "February 2026"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "IMOAnswerBench"
score = 91.0
metric = "accuracy"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "GPQA Diamond"
score = 91.2
metric = "accuracy"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "NL2Repo"
score = 48.9
metric = "resolve rate"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "DeepSWE"
score = 46.2
metric = "resolve rate"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "Program Bench"
score = 63.7
metric = "score"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "Terminal-Bench"
score = 81.0
metric = "success rate"
harness = "Terminus 2"
version = "2.1"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "PostTrainBench"
score = 34.3
metric = "score"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "SWE Marathon"
score = 13.0
metric = "resolve rate"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "MCP Atlas"
score = 76.8
metric = "score"
dataset = "public subset"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"

[[benchmarks]]
name = "Tool-Decathlon"
score = 48.2
metric = "score"
source = "https://z.ai/blog/glm-5.2"
date = "2026-06-16"