Inference
EN
Z.AI

GLM-5.3

GLM-5.3 is a reasoning model from Z.AI in the GLM-5 family. 63 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-08
RankedEnough results for an index.
IndexAll categories, out of 100.65.8±3.5Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.86/sMedian 67/s
First tokenSeconds to the first output token.0.3 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.07batch $0.322Median $0.60
Output / 1MUS dollars per 1M output tokens.$1.40batch $0.989 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1476±5 #25Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
z-ai/glm-5.3
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.
$0.069
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.
22 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
OpenRouterRouter$0.07$1.401.05M
CoreWeaveHost$0.75$3.001.05M
VultrCloud$0.75$3.001.05M
Fireworks AIHost$0.98$3.081.05M
FriendliHost$1.26$3.961.05M
Deep InfraHost$0.90$4.001.05M
Zhipu AIMaker$1.20$4.181M
Arcee AIHost$1.40$4.401M
BasetenHost$1.40$4.401.05M
Cloudflare Workers AICloud$1.40$4.401.05M
DigitalOceanCloud$1.40$4.401.05M
MistralHost$1.40$4.401.05M
ModalHost$1.40$4.401M
NebiusHost$1.40$4.401.02M
Novita AIHost$1.40$4.401.05M
SiliconFlowHost$1.40$4.401.05M
Together AIHost$1.40$4.401.05M
Z.AIMaker$1.40$4.401M
Amazon BedrockCloud$1.68$5.281M
TinfoilHost$1.80$5.751.05M
ScalewayCloud$2.01$6.15256K
PrivatemodeHost$1.79$6.631M

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.28 counted68.233 / 218
CodingCode writing and repair.17 counted65.738 / 209
ReasoningLogic problems and puzzles.6 counted64.546 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.6 counted59.366 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted66.8—
MathMath problems.5 counted63.138 / 168

Results

63 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
SWE-benchCoding95.4% ±0.95 / 87
OpenHarmony Bench v1.0Coding60.8%1 / 13
GDPval-AA normalizedAgentic57.5%12 / 133
Artificial Analysis SciCodeCoding59.0%12 / 114
Terminal-Bench 2.1 · BenchLMAgentic88.2%10 / 86
Artificial Analysis GPQA DiamondKnowledge91.7%24 / 199
Artificial Analysis Tau3-BankingAgentic50.3%3 / 22
Critical Physics TasksReasoning19.1%32 / 214
GPQA diamond · Epoch AIKnowledge90.9% ±1.635 / 217
Artificial Analysis Humanity's Last ExamKnowledge42.3%39 / 222
Humanity's Last Exam with toolsAgentic62.5%7 / 36
Artificial Analysis Long Context ReasoningReasoning79.7%45 / 217
NL2RepoCoding58.0%8 / 36
LiveBench Agentic CodingAgentic60.9%15 / 66
GPQA Diamond · Vals AIKnowledge88.1% ±1.935 / 135
Terminal-Bench 4.0Agentic38.9% ±1.817 / 61
MMLU ProKnowledge86.8% ±0.338 / 135
IOICoding68.4% ±7.513 / 45
OTIS Mock AIME 2024-2025Math91.1% ±3.558 / 200
Vibe Code Bench v1.1Coding78.1% ±3.936 / 123
CyberGymAgentic84.5%9 / 29
Code MigrationCoding44.2% ±4.324 / 77
FrontierMath-Tiers-1-3-v2-PrivateMath68.8% ±2.731 / 90
Terminal-Bench 2.1 · Vals AIAgentic71.5% ±2.026 / 75
Mystery Game PuzzlesReasoning33.0% ±4.728 / 80
LiveBench CodingCoding79.0%24 / 66
Terminal-Bench 4.0.0Agentic41.8% ±1.69 / 23
Terminal-Bench 4.1.0Agentic41.2% ±1.79 / 23
LiveBench DesignAgentic77.3%21 / 52
Chess PuzzlesReasoning37.0% ±4.939 / 90
LiveBench ReasoningReasoning85.8%35 / 80
Hermes Index: agentic capability and cost per task inside Hermes AgentAgentic39.2%7 / 16
LiveBench LanguageKnowledge79.9%29 / 66
Agent Arena task outcomeAgentic3.124 / 54
Artificial Analysis Omniscience AccuracyKnowledge32.3%73 / 164
Vibe Code Bench 1-100Coding20.0% ±3.812 / 26
LiveBench Instruction FollowingInstruction69.3%31 / 66
LiveCodeBenchCoding80.5% ±1.167 / 142
Artificial Analysis DeepsecBench-AAAgentic24.3%10 / 21
Artificial Analysis AutomationBenchAgentic62.2%11 / 23
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
LiveBench ResearchAgentic73.2%26 / 52
AutomationBenchAgentic48.2%16 / 32
Artificial Analysis ITBench-AAAgentic46.1%12 / 24
FrontierMath-Tier-4-v2-PrivateMath29.3% ±7.236 / 72
DeepSWEAgentic66.9%31 / 59
Artificial Analysis CWE-Bench-AAAgentic55.8%10 / 19
Artificial Analysis CyberGym-E2E-AAAgentic29.0%10 / 19
LiveBench Coding 2.0Coding82.8%28 / 52
Artificial Analysis Terminal-Bench v4.0Agentic41.9%12 / 22
FrontierSWE v2Coding30.2%12 / 22
LiveBench Long AgenticAgentic35.9%28 / 51
SimpleQA VerifiedKnowledge41.0% ±1.651 / 91
cursorBench40Coding42.6%12 / 21
Artificial Analysis Cyber IndexAgentic36.4%11 / 19
Agent Arena steerabilityAgentic−1.232 / 54
Toolathlon-VerifiedAgentic73.0%14 / 23
LiveBench MathematicsMath87.9%41 / 66
ProofBench v1.1Math49.0% ±5.034 / 54
Artificial Analysis Terminal-Bench v2.1Agentic83.9%14 / 21
Agent Arena command recoveryAgentic−0.134 / 50
LiveBench Data AnalysisReasoning70.2%56 / 80
SkillsBenchCoding47.5% ±4.525 / 35
Agents' Last ExamAgentic28.5%15 / 21
ExploitGymAgentic15.0%17 / 23
Bug Hunt BenchCoding19.0%21 / 25
VulcanBench v3Coding78.3%12 / 14
Medical Long Context Reasoning (MLCR-AA)Reasoning17.2%19 / 22
Artificial Analysis AnalystAgentAgentic38.8%23 / 26
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding19.0%18 / 19
CWE-bench v2Agentic42.0%11 / 11
Artificial Analysis GDP.pdfAgentic11.2%22 / 22
Tax Agent BenchFinance43.3% ±3.67 / 82
MedScribeHealthcare88.8% ±2.011 / 109
Legal Research BenchLegal49.0% ±3.510 / 90
LegalBenchLegal84.8% ±0.429 / 149
Public Benefits Bench v1.1Public benefits68.5% ±1.210 / 50
ProgramBenchCoding1.5% ±0.914 / 65
Finance Agent (v2)Finance55.8% ±2.122 / 79
HLAB-Vals-1.1Legal14.6% ±2.912 / 35
CyberBench v1.1Other72.1% ±5.418 / 48
SRE BenchOther9.2% ±1.815 / 36
TaxEval v2Finance72.4% ±0.962 / 143
MedCodeHealthcare42.9% ±2.149 / 106
Terminal-Bench-ScienceScience5.7% ±2.822 / 44
EMBFinance56.3% ±3.346 / 75
MysteryMechanismScience23.0% ±2.820 / 30
b3: Backbone Breaker BenchmarkAgentic93.0%—
Refusal of harmful cyber requests (JailbreakBench, StrongREJECT, AgentHarm)Agentic87.0%—
FrontierSWECoding78.1%—
Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindAgentic71.8%—
Finch (FinWorkBench): Finance & Accounting across Spreadsheet-Centric Enterprise WorkflowsAgentic65.1%—
SWE-MarathonCoding42.5%—
PostTrain BenchCoding39.8%—
Artificial Analysis EnterpriseOps-GymAgentic36.4%—
terminalBench3Agentic28.3%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYLiveBenchNo licence statedTerminal-BenchNo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index67.2 → 65.8
23 Oct 202612 Nov 2026
Output / 1M$2.50 → $1.40
23 Oct 202612 Nov 2026

Questions

How much does GLM-5.3 cost?
$0.07 per 1M input tokens, $1.40 per 1M output tokens.
How good is GLM-5.3?
Its index is 65.8 out of 100. That is place 39 of 283 ranked models.
What is GLM-5.3 best at?
Agentic: 68.2 out of 100, place 33 of 218.
How fast is GLM-5.3?
86 output tokens per second, 0.3 s to the first token.
What is the context window of GLM-5.3?
1.05M tokens.
Is GLM-5.3 open source?
Yes. Its weights are published.
When was GLM-5.3 released?
August 2026.

Badge

GLM-5.3 on Inference360: place 39 of 283, index 65.8
HTML
<a href="https://inference360.ai/models/glm-5-3"><img src="https://inference360.ai/badges/glm-5-3.svg" alt="GLM-5.3 on Inference360: place 39 of 283, index 65.8" height="20"></a>
Markdown
[![GLM-5.3 on Inference360: place 39 of 283, index 65.8](https://inference360.ai/badges/glm-5-3.svg)](https://inference360.ai/models/glm-5-3)

Compare with

Qwen3.8 Max65.6GPT-6 Luna65.5Claude Haiku 5.566.3dots3-note Preview63.4GLM-5.3-Flash63.1

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 LatestUnrankedGLM-5.3-Flash63.1GLM Flash LatestUnrankedGLM-5.3 FlashXUnrankedGLM-5.3 PrimeUnranked