Inference
EN
Z.AI

GLM-5.2

GLM-5.2 is a reasoning model from Z.AI in the GLM-5 family. 50 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-06
RankedEnough results for an index.
IndexAll categories, out of 100.60.6±3.8Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.111/sMedian 67/s
First tokenSeconds to the first output token.0.6 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.<$0.01batch $0.322Median $0.60
Output / 1MUS dollars per 1M output tokens.$6.80batch $0.989 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1469±4 #32Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
z-ai/glm-5.2
Max outputMaximum output tokens per request.
944K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
<$0.01
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
26 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
Deep InfraHost$0.75$2.401.05M
CoreWeaveHost$0.62$2.421.05M
VultrCloud$0.75$3.001.05M
GMI CloudHost$0.979$3.081M
ChutesHost$1.25$3.951.05M
SiliconFlowHost$1.30$4.091.05M
Volcengine ArkCloud$1.19$4.161M
Zhipu AIMaker$1.20$4.181M
Alibaba CloudHost$1.40$4.401M
Arcee AIHost$1.40$4.40262K
BasetenHost$1.40$4.401.05M
Cloudflare Workers AIDeprecatedCloud$1.40$4.40262K
CrusoeDeprecatedHost$1.40$4.401M
DatabricksCloud$1.40$4.401M
DigitalOceanCloud$1.40$4.40262K
FriendliHost$1.40$4.401.05M
Google Vertex AICloud$1.40$4.401M
MistralDeprecatedHost$1.40$4.401.05M
NebiusHost$1.40$4.401.05M
Novita AIHost$1.40$4.401.05M
Together AIHost$1.40$4.401.05M
Z.AIMaker$1.40$4.401M
Berget AIDeprecatedHost$1.54$4.84524K
RegoloHost$2.31$6.0096K
ScalewayCloud$2.02$6.17256K
OpenRouterRouter<$0.01$6.801.05M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

753.33B parameters
16-bit
1583.0 GB
8-bit
841.4 GB
4-bit
480.5 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.13 counted61.761 / 218
CodingCode writing and repair.14 counted63.550 / 209
ReasoningLogic problems and puzzles.8 counted58.673 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.6 counted55.790 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted55.126 / 63
MathMath problems.7 counted59.651 / 168

Results

50 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
τ²-Bench Tool-Agent-User EvaluationAgentic99.1%1 / 149
AIME 2026Math99.2%1 / 34
Critical Physics TasksReasoningAbove fitFar above this model's index.20.9%26 / 214
SWE-Bench verified · Epoch AICoding78.7% ±1.94 / 32
GPQA diamond · Epoch AIKnowledge91.9% ±1.629 / 217
Graduate-Level Google-Proof Q&AKnowledge91.2%23 / 141
Artificial Analysis Humanity's Last ExamKnowledge41.1%42 / 222
Artificial Analysis IFBenchInstruction73.3%29 / 146
SWE-benchCoding82.8% ±1.718 / 87
Artificial Analysis GPQA DiamondKnowledge89.5%43 / 199
Harvard-MIT Mathematics Tournament November 2025Math94.4%3 / 12
GPQA Diamond · BenchLMKnowledge91.2%22 / 84
APEX-Agents-AAAgentic33.7%8 / 30
MMLU ProKnowledge86.7% ±0.339 / 135
Artificial Analysis Long Context ReasoningReasoning78.3%63 / 217
Harvard-MIT Mathematics Tournament February 2026Math92.5%8 / 27
SWE-bench ProCoding62.1%26 / 83
GDPval-AA normalizedAgentic43.6%42 / 133
LiveBench CodingCoding79.7%21 / 66
GPQA Diamond · Vals AIKnowledge85.6% ±1.845 / 135
NL2RepoCoding48.9%13 / 36
OTIS Mock AIME 2024-2025Math86.4% ±4.273 / 200
MCP AtlasAgentic76.8%19 / 52
Terminal-Bench 2.1 · BenchLMAgentic81.0%32 / 86
OpenHarmony Bench v1.0Coding58.4%5 / 13
Terminal-Bench HardAgentic50.8%9 / 23
Artificial Analysis SciCodeCoding51.2%47 / 114
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding63.7%8 / 19
Terminal-Bench 2.1 · Vals AIAgentic67.8% ±1.032 / 75
Code MigrationCoding37.9% ±4.133 / 77
τ²-bench BankingAgentic37.1%12 / 28
ARC-AGI-1 (semi-private)Reasoning77.0%55 / 126
Humanity's Last ExamKnowledge40.5%34 / 77
ARC-AGI-2 (semi-private)Reasoning22.8%59 / 131
Vibe Code Bench v1.1Coding64.0% ±4.856 / 123
FrontierMath-Tiers-1-3-v2-PrivateMath59.2% ±3.043 / 90
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
FrontierMath-Tier-4-v2-PrivateMath29.3% ±7.236 / 72
Agent Arena task outcomeAgentic1.727 / 54
Humanity's Last Exam without toolsKnowledge40.5%27 / 53
LiveBench MathematicsMath89.8%34 / 66
ToolathlonAgentic48.2%12 / 22
LiveBench Data AnalysisReasoning73.7%44 / 80
Agent Arena command recoveryAgentic0.728 / 50
Agent Arena steerabilityAgentic−1.031 / 54
LiveBench Agentic CodingAgentic51.8%38 / 66
Humanity's Last Exam with toolsAgentic54.7%21 / 36
Artificial Analysis ITBench-AAAgentic42.7%14 / 24
Chess PuzzlesReasoning22.0% ±4.257 / 90
MirrorCodeCoding36.7% ±11.525 / 39
LiveCodeBenchCoding69.5% ±1.292 / 142
Artificial Analysis Omniscience AccuracyKnowledge24.3%108 / 164
SimpleQA VerifiedKnowledge34.2% ±1.560 / 91
LiveBench LanguageKnowledge76.2%44 / 66
PostTrainBench v1.1Coding31.7%10 / 15
Mystery Game PuzzlesReasoning15.0% ±3.659 / 80
SkillsBenchCoding45.1% ±4.427 / 35
LiveBench ReasoningReasoning78.6%62 / 80
LiveBench Instruction FollowingInstruction62.3%53 / 66
cursorBench32Coding55.0%15 / 18
EBR-benchReasoning9.5%26 / 31
Terminal-Bench 3.0Agentic4.6%13 / 13
CorpFin v2Finance66.1% ±0.921 / 131
LegalBenchLegal84.1% ±0.539 / 149
ProgramBenchCoding0.5% ±0.519 / 65
TaxEval v2Finance73.3% ±0.950 / 143
MedScribeHealthcare83.5% ±2.046 / 109
ARC-AGI-2 (public eval)Reasoning20.8%53 / 122
ARC-AGI-1 (public eval)Reasoning80.4%51 / 116
EMBFinance61.5% ±2.933 / 75
MedCodeHealthcare40.8% ±2.260 / 106
Legal Research BenchLegal31.3% ±3.252 / 90
Finance Agent (v2)Finance49.7% ±0.946 / 79
SRE BenchOther0.0%33 / 36
MMAnswerBenchMath91.0%—
Artificial Analysis EnterpriseOps-GymAgentic42.7%—
ResearchClawBenchAgentic20.7%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYLiveBenchNo licence statedVals AINo licence statedARC Prize FoundationTerms restrict reuseSierra ResearchMITLMArenaCC BY 4.0models.devMIT

History

5 releases
Index61.2 → 60.6
23 Oct 202612 Nov 2026
Output / 1M$2.42 → $6.80
23 Oct 202612 Nov 2026

Questions

How much does GLM-5.2 cost?
<$0.01 per 1M input tokens, $6.80 per 1M output tokens. A free tier is also available.
How good is GLM-5.2?
Its index is 60.6 out of 100. That is place 68 of 283 ranked models.
What is GLM-5.2 best at?
Coding: 63.5 out of 100, place 50 of 209.
How fast is GLM-5.2?
111 output tokens per second, 0.6 s to the first token.
What is the context window of GLM-5.2?
1.05M tokens.
Is GLM-5.2 open source?
Yes. Its weights are published.
When was GLM-5.2 released?
June 2026.

Badge

GLM-5.2 on Inference360: place 68 of 283, index 60.6
HTML
<a href="https://inference360.ai/models/glm-5-2"><img src="https://inference360.ai/badges/glm-5-2.svg" alt="GLM-5.2 on Inference360: place 68 of 283, index 60.6" height="20"></a>
Markdown
[![GLM-5.2 on Inference360: place 68 of 283, index 60.6](https://inference360.ai/badges/glm-5-2.svg)](https://inference360.ai/models/glm-5-2)

Compare with

Qwen3.7 Max60.5Sage-A1.560.5Mistral Large 460.7Gemini 3.1 Pro Preview61.5dots3-note Preview63.4

Lower price

dots3-note Preview63.4 · FreeGLM-5.3-Flash63.1 · $0.078/1MLing 3.1 Flash67.2 · $0.135/1M

More from Z.AI

All models
GLM-5.365.8GLM LatestUnrankedGLM-5.156.0GLM-5.3-Flash63.1GLM Flash LatestUnranked