Inference
EN
Z.AI

GLM-5.3-Flash

GLM-5.3-Flash is a reasoning model from Z.AI in the GLM-5 family. 56 benchmarks count toward its index, in 8 categories.

  • BudgetTop-25% index, best per paid dollar.
  • Open weights
  • Reasoning
  • Input: text · image · video
  • Released 2026-08
RankedEnough results for an index.
IndexAll categories, out of 100.63.1±2.2Median 51.0
CoverageIndex weight that has results.100%
SpeedOutput tokens per second.122/sMedian 67/s
First tokenSeconds to the first output token.0.9 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.04Median $0.60
Output / 1MUS dollars per 1M output tokens.$0.078Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1471±4 #31Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
z-ai/glm-5.3-flash
Max outputMaximum output tokens per request.
131K
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.
$0.03
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.
21 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
OpenRouterRouter$0.04$0.0781.05M
VultrCloud$0.10$0.351.05M
CoreWeaveHost$0.12$0.381.05M
Volcengine ArkCloud$0.119$0.4161M
Zhipu AIMaker$0.12$0.4181M
Arcee AIHost$0.15$0.501M
BasetenHost$0.15$0.501.05M
Cloudflare Workers AIDeprecatedCloud$0.15$0.501.05M
Deep InfraHost$0.15$0.501.05M
DigitalOceanCloud$0.15$0.501.05M
Fireworks AIHost$0.15$0.501.05M
FriendliHost$0.15$0.501.05M
MistralHost$0.15$0.501.05M
ModalHost$0.15$0.501M
NebiusHost$0.15$0.501.02M
Novita AIHost$0.15$0.501.05M
SiliconFlowHost$0.15$0.501.05M
Together AIHost$0.15$0.501.05M
Z.AIMaker$0.15$0.501M
Berget AIHost$0.29$0.581.05M
PrivatemodeHost$0.231$0.7511M

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

Run locally

321.32B parameters
16-bit
675.8 GB
8-bit
359.5 GB
4-bit
205.5 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.21 counted69.726 / 218
CodingCode writing and repair.14 counted62.356 / 209
ReasoningLogic problems and puzzles.5 counted58.972 / 227
MultimodalTasks with images and text.6 counted65.025 / 84
KnowledgeFacts and expert knowledge.3 counted57.776 / 268
MultilingualTasks in many languages.1 counted64.0—
InstructionTasks with strict prompt rules.1 counted33.5—
MathMath problems.5 counted57.356 / 168

Results

56 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
EuroEval ItalianMultilingual72.2% ±0.36 / 174
EuroEval PortugueseMultilingual69.9% ±0.47 / 182
EuroEval FrenchMultilingual71.1% ±0.48 / 178
EuroEval DutchMultilingual66.0% ±0.48 / 175
EuroEval GermanMultilingual59.8% ±0.69 / 191
EuroEval SwedishMultilingual69.3% ±0.710 / 195
EuroEval SpanishMultilingual62.2% ±0.511 / 176
Toolathlon-VerifiedAgentic78.4%2 / 23
EuroEval PolishMultilingual61.0% ±0.419 / 209
SWE-benchCoding92.0% ±1.210 / 87
Artificial Analysis ITBench-AAAgentic51.2%4 / 24
GPQA diamond · Epoch AIKnowledge90.2% ±1.740 / 217
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal70.5%8 / 39
CharXiv ReasoningMultimodal89.4%11 / 49
LiveBench Browser UseAgentic84.0%11 / 49
Artificial Analysis Tau3-BankingAgentic47.2%5 / 22
OTIS Mock AIME 2024-2025Math93.9% ±2.249 / 200
MMMU ProMultimodal86.0% ±0.823 / 92
NL2RepoCoding56.3%9 / 36
ARC-AGI-2 (semi-private)Reasoning65.8%33 / 131
Terminal-Bench 2.1 · BenchLMAgentic84.3%22 / 86
ARC-AGI-1 (semi-private)Reasoning91.0%33 / 126
Artificial Analysis CyberGym-E2E-AAAgenticAbove fitFar above this model's index.74.0%6 / 19
Artificial Analysis Cyber IndexAgentic49.9%6 / 19
GPQA Diamond · Vals AIKnowledge86.4% ±2.143 / 135
Terminal-Bench 4.0Agentic37.4% ±2.720 / 61
MMLU ProKnowledge86.1% ±0.349 / 135
LiveBench CodingCoding79.0%24 / 66
LiveBench Agentic CodingAgentic56.8%25 / 66
LiveBench DesignAgentic77.2%22 / 52
LiveBench Data AnalysisReasoning76.4%35 / 80
OpenHarmony Bench v1.0Coding57.3%6 / 13
AutomationBenchAgentic48.8%15 / 32
LiveCodeBenchCoding80.5% ±1.167 / 142
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
Terminal-Bench 2.1 · Vals AIAgentic62.9% ±2.238 / 75
IOICoding52.5% ±2.623 / 45
Terminal-Bench 4.0.0Agentic35.8% ±1.812 / 23
Terminal-Bench 4.1.0Agentic35.5% ±1.812 / 23
Artificial Analysis CWE-Bench-AAAgentic55.8%10 / 19
Humanity's Last Exam with toolsAgentic55.3%19 / 36
Agent Arena task outcomeAgentic1.329 / 54
FrontierMath-Tiers-1-3-v2-PrivateMath55.8% ±2.949 / 90
Artificial Analysis Terminal-Bench v2.1Agentic84.3%12 / 21
LiveBench Coding 2.0Coding80.6%31 / 52
LiveBench ResearchAgentic70.1%32 / 52
Artificial Analysis DeepsecBench-AAAgentic19.9%13 / 21
LiveBench LanguageKnowledge77.3%41 / 66
OfficeQA ProMultimodal62.4%10 / 16
DeepSWEAgentic63.4%37 / 59
LiveBench Long AgenticAgentic28.4%32 / 51
Chess PuzzlesReasoning21.0% ±4.159 / 90
Vibe Code Bench v1.1Coding30.8% ±5.382 / 123
Code MigrationCoding20.5% ±4.252 / 77
Vibe Code Bench 1-100Coding16.0% ±3.318 / 26
FrontierMath-Tier-4-v2-PrivateMath17.1% ±5.952 / 72
Agent Arena steerabilityAgentic−5.039 / 54
BabyVisionMultimodal53.4%8 / 11
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal80.8%19 / 26
Hermes Index: agentic capability and cost per task inside Hermes AgentAgentic35.0%12 / 16
Multimodal Knowledge Agent BenchmarkAgentic36.2%15 / 19
ProofBench v1.1Math21.0% ±4.143 / 54
Agents' Last ExamAgentic26.3%17 / 21
LiveBench ReasoningReasoning77.6%65 / 80
Agent Arena command recoveryAgentic−5.741 / 50
Mystery Game PuzzlesReasoning8.0% ±2.766 / 80
SkillsBenchCoding40.2% ±4.430 / 35
cursorBench40Coding36.8%18 / 21
LiveBench MathematicsMath81.2%57 / 66
FrontierSWE v2Coding18.1%19 / 22
Bug Hunt BenchCoding17.7%24 / 25
LiveBench Instruction FollowingInstruction52.8%66 / 66
MedScribeHealthcare88.9% ±1.99 / 109
TaxEval v2Finance75.6% ±0.814 / 143
Finance Agent (v2)Finance57.9% ±2.113 / 79
SAGEEducation89.3% ±1.85 / 29
ARC-AGI-2 (public eval)Reasoning66.8%29 / 122
Legal Research BenchLegal45.2% ±3.522 / 90
LegalBenchLegal83.9% ±0.445 / 149
Tax Agent BenchFinance31.6% ±3.428 / 82
ARC-AGI-1 (public eval)Reasoning92.5%40 / 116
ProgramBenchCoding0.0%34 / 65
HLAB-Vals-1.1Legal10.4% ±2.620 / 35
CyberBench v1.1Other67.9% ±5.628 / 48
EMBFinance55.9% ±3.648 / 75
Terminal-Bench-ScienceScience2.9% ±2.033 / 44
SRE BenchOther0.8% ±0.528 / 36
Multimodal Multi-disciplinary Video UnderstandingMultimodal80.5%—
Chartography with image and code toolsMultimodal78.0%—
Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindAgentic68.9%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYVals AINo licence statedARC Prize FoundationTerms restrict reuseLiveBenchNo licence statedEuroEvalMITTerminal-BenchNo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index63.7 → 63.1
23 Oct 202612 Nov 2026
Output / 1M$0.20 → $0.078
23 Oct 202612 Nov 2026

Questions

How much does GLM-5.3-Flash cost?
$0.04 per 1M input tokens, $0.078 per 1M output tokens.
How good is GLM-5.3-Flash?
Its index is 63.1 out of 100. That is place 60 of 283 ranked models.
What is GLM-5.3-Flash best at?
Agentic: 69.7 out of 100, place 26 of 218.
How fast is GLM-5.3-Flash?
122 output tokens per second, 0.9 s to the first token.
What is the context window of GLM-5.3-Flash?
1.05M tokens.
Is GLM-5.3-Flash open source?
Yes. Its weights are published.
When was GLM-5.3-Flash released?
August 2026.

Badge

GLM-5.3-Flash on Inference360: place 60 of 283, index 63.1
HTML
<a href="https://inference360.ai/models/glm-5-3-flash"><img src="https://inference360.ai/badges/glm-5-3-flash.svg" alt="GLM-5.3-Flash on Inference360: place 60 of 283, index 63.1" height="20"></a>
Markdown
[![GLM-5.3-Flash on Inference360: place 60 of 283, index 63.1](https://inference360.ai/badges/glm-5-3-flash.svg)](https://inference360.ai/models/glm-5-3-flash)

Compare with

Claude Opus 4.7 (Adaptive)63.2Qwen3.8 Max (0902)63.4dots3-note Preview63.4Qwen3.8-Omni-Flash62.7Gemini 3.1 Pro62.3

Lower price

dots3-note Preview63.4 · Free

More from Z.AI

All models
GLM Flash LatestUnrankedGLM LatestUnrankedGLM-5.365.8GLM-5.3 FlashXUnrankedGLM-5.3 PrimeUnranked