Zhipu AI Open weights: zai-org/GLM-5.3Reasoning modelCompare with another model

GLM 5.3

Reasoning effort

GLM 5.3 is a Zhipu AI open-weights reasoning model, released 18 Aug 2026. Its best configuration (max reasoning effort) ranks #59 of 775 on the BenchLeader Index at 63.6 ±3.2, in the upper half. The ± is the point: 140 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in composite (80) and lowest in agents & tools (58). At $2.15 per million tokens blended it is pricier than most ranked models. Output speed of 84 tokens per second puts it slower than most, with a first answer in 26.4 s. It has been measured at 4 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 19 Oct 2026.

Blended price: $2.15/M; $1.40 in · $4.40 out
Output speed: 84 tok/s; measured by Artificial Analysis
First answer: 26 s; first token 1.19 s
Context: 1M
Full answer: 32 s; median, reasoning included
Cost per run: $2.01; one full Intelligence Index run
Released: 18 Aug 2026

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index64
Reasoning69
Coding62
Agents & tools58
Maths60
Knowledge59
Human preference67
Long context65
Composite80

Versions

Zhipu AI has shipped 8 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
GLM 5.3maxthis page	18 Aug 2026	63.6	#59
GLM 5.2max	16 Jun 2026	62.4	#85
GLM 5.1thinking	7 Apr 2026	56.6	#197
GLM-5thinking	11 Feb 2026	58.9	#141
GLM 4.7thinking	22 Dec 2025	52.1	#324
GLM-4.6	30 Sept 2025	49.9	#379
GLM-4.5	28 Jul 2025	51.8	#332
GLM-4-0520	–	–	–

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
low	55.6	#225	86 tok/s	26 s	$0.0019	Coding 43 · Composite 68 · Knowledge 58 · Long context 61 · Reasoning 65
high	–	–	84 tok/s	1.19 s	$0.0019
maxbest	63.6	#59	84 tok/s	26 s	$0.0019	Agents & tools 58 · Coding 62 · Composite 80 · Human preference 67 · Knowledge 59 · Long context 65 · Maths 60 · Reasoning 69
not stated	–	–	84 tok/s	1.19 s	$0.0019	Agents & tools 62 · Composite 42

What more thinking costs

Turning the effort up buys index points and multiplies the bill. Here max costs 2.4× low for +8.0 on the index — 6.5 points per doubling of spend. Compare that with other models.

How its index has moved

8 Sept 2026 to 19 Oct 2026

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Metadata

Benchmark	max	Source	Trend
AA Openness Indexnot in index	33.3#32	Artificial Analysis

Reasoning

Benchmark	low	max	not stated	Source
GPQA Diamond	–	90.9%#34	–	Epoch AI Benchmarking Hub
Arena Hard Prompts	–	1503#27	–	Arena (LMArena)
LiveBench 2.0 Reasoningnot in indexeffort not stated	–	–	85.8%#28	LiveBench
LiveBench 1.0 Reasoningnot in indexeffort not stated	–	–	85.8%#35	LiveBench
GPQA Diamond (AA)not in index	–	91.7%#45	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	36.6%#109	42.3%#71	–	Artificial Analysis
GPQA Diamond (Vals)not in index	–	88.1%#34	–	Vals AI
CritPt	14.6%#96	19.1%#65	–	Artificial Analysis
MysteryMechanismnot in index	–	23.0%#18	–	Vals AI
Chess Puzzlesnot in index	–	37.0%#43	–	Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index	–	33.0%#27	–	Epoch AI Benchmarking Hub
LMCAnot in indexeffort not stated	–	–	55.5%#45	Epoch AI Benchmarking Hub
DTBenchnot in indexeffort not stated	–	–	87.7%#99	Epoch AI Benchmarking Hub
BTF-3not in indexeffort not stated	–	–	14.9%#1	Epoch AI Benchmarking Hub

Coding

Benchmark	low	high	max	not stated	Source
SciCode	42.0%#151	–	59.0%#15	–	SciCode
WeirdML	–	–	75.4%#29	–	WeirdML
FrontierCode	–	–	40.1%#26	–	Cognition
Arena Coding	–	–	1519#38	–	Arena (LMArena)
Arena WebDev	–	–	1621#23	–	Arena (LMArena)
LiveBench 2.0 Codingnot in indexeffort not stated	–	–	–	82.8%#27	LiveBench
LiveBench 1.0 Codingnot in indexeffort not stated	–	–	–	79.0%#26	LiveBench
SciCode (AA)not in index	42.0%#160	–	59.0%#16	–	Artificial Analysis
LiveCodeBench	–	–	80.5%#65	–	Vals AI
IOI	–	–	68.4%#13	–	Vals AI
SWE-bench (Vals)not in index	–	–	95.4%#5	–	Vals AI
Code Migrationnot in index	–	–	44.2%#23	–	Vals AI
ProgramBenchnot in index	–	–	1.5%#13	–	Vals AI
Vibe Code Bench 1-100not in index	–	–	20.0%#12	–	Vals AI
Vibe Code Bench v1.1not in index	–	–	78.1%#30	–	Vals AI
DeepSWE v1.1not in index	–	–	69.0%#16	–	Epoch AI Benchmarking Hub
CursorBenchnot in index	33.3%#57	38.0%#42	42.6%#30	–	Cursor
ALE-Benchnot in index	–	1317.4#31	–	–	Epoch AI Benchmarking Hub
FrontierSWEnot in index	–	–	30.2%#10	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	34.9%#34	–	41.9%#28	–	Artificial Analysis
Terminal-Bench 2.1 (AA)not in index	–	–	83.9%#34	–	Artificial Analysis
Startrise Frontendnot in indexeffort not stated	–	–	–	73.2%#12	Startrise
FrontierCode 1.1 (AA)not in index	–	–	40.1%#26	–	Artificial Analysis

Agents & tools

Benchmark	low	max	not stated	Source
Terminal-Bench	–	41.8%#47	–	Terminal-Bench
APEX-Agentseffort not stated	–	–	56.6%#16	Mercor
Arena Agent	–	1.3#25	–	Arena (LMArena)
LiveBench Long Agenticnot in indexeffort not stated	–	–	35.9%#28	LiveBench
LiveBench Browsernot in indexeffort not stated	–	–	0.0%#45	LiveBench
LiveBench Agentic Codingnot in indexeffort not stated	–	–	60.9%#15	LiveBench
Terminal-Bench 2.1 (Vals)	–	71.5%#26	–	Vals AI
MCP Atlaseffort not stated	–	–	84.2%#5	Scale AI SEAL
HiL-Bencheffort not stated	–	–	58.3%#9	Scale AI SEAL
GDPval-AA v2.1not in index	41.1%#99	57.5%#18	–	Artificial Analysis
τ³-Banking (AA)	–	50.3%#4	–	Artificial Analysis
ITBench SRE (AA)not in index	–	46.1%#12	–	Artificial Analysis
Analyst Agent (AA)not in index	–	38.8%#24	–	Artificial Analysis
CyberBenchnot in index	–	72.1%#17	–	Vals AI
Finance Agent v2not in index	–	55.8%#22	–	Vals AI
Harvey's Legal Agent Benchmarknot in index	–	14.6%#6	–	Vals AI
Legal Research Benchnot in index	–	49.0%#9	–	Vals AI
SkillsBenchnot in index	–	47.5%#24	–	Vals AI
SREBenchnot in index	–	9.2%#15	–	Vals AI
Tax Agent Benchnot in index	–	39.7%#9	–	Vals AI
Terminal-Bench 4.0 (Vals)not in index	–	38.9%#12	–	Vals AI
Terminal-Bench Sciencenot in index	–	5.7%#21	–	Vals AI
Vending-Bench 2not in indexeffort not stated	–	–	8163.6#13	Andon Labs
AutomationBenchnot in index	52.0%#72	62.2%#26	–	Artificial Analysis
GDP.pdf (AA)not in index	8.8%#136	11.2%#117	–	Artificial Analysis
Harvey LAB-AA v1.1not in index	–	0.3%#22	–	Harvey
EnterpriseOps-Gymnot in index	–	36.4%#32	–	Artificial Analysis
AA-Briefcase v1.1not in index	–	1507#28	–	Artificial Analysis
Terminal-Bench-Science 0.1 (AA)not in index	–	9.5%#28	–	Artificial Analysis
CWE-Bench (AA)	–	55.8%#8	–	Artificial Analysis
DeepSec-Bench (AA)	–	24.3%#8	–	Artificial Analysis
CyberGym end-to-end (AA)	–	29.0%#9	–	Artificial Analysis

Maths

Benchmark	max	not stated	Source
FrontierMath Tiers 1–3	68.8%#29	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	29.3%#39	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	91.1%#69	–	Epoch AI Benchmarking Hub
ProofBench	49.0%#33	–	Vals AI
LiveBench Mathematicsnot in indexeffort not stated	–	87.9%#44	LiveBench
Arena Maths	1494#23	–	Arena (LMArena)

Knowledge

Benchmark	low	max	not stated	Source
SimpleQA Verified	–	41.0%#45	–	Epoch AI Benchmarking Hub
LiveBench 2.0 Data Analysisnot in indexeffort not stated	–	–	70.2%#35	LiveBench
LiveBench Researchnot in indexeffort not stated	–	–	73.2%#24	LiveBench
LiveBench 1.0 Data Analysisnot in indexeffort not stated	–	–	70.2%#51	LiveBench
AA-Omniscience	-8.4#179	14.3#84	–	Artificial Analysis
MMLU-Pro	–	86.8%#37	–	Vals AI
LegalBench	–	84.8%#29	–	Vals AI
TaxEval	–	72.4%#64	–	Vals AI
Excel Modeling Benchmarknot in index	–	56.3%#45	–	Vals AI
MedCodenot in index	–	42.9%#48	–	Vals AI
MedScribenot in index	–	88.8%#11	–	Vals AI
Public Benefits Benchnot in index	–	68.5%#9	–	Vals AI
AA-Omniscience: accuracynot in index	33.9%#158	33.9%#159	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	36.0%#183	70.5%#40	–	Artificial Analysis

Instruction following

Benchmark	max	not stated	Source
LiveBench Languagenot in indexeffort not stated	–	79.9%#31	LiveBench
LiveBench Instruction Followingnot in indexeffort not stated	–	69.3%#32	LiveBench
Arena Instruction Followingnot in index	1475#26	–	Arena (LMArena)

Human preference

Benchmark	max	Source
Arena Text	1479#28	Arena (LMArena)
Arena Creative Writingnot in index	1453#35	Arena (LMArena)
Arena Multi-turnnot in index	1480#41	Arena (LMArena)
Arena Longer Queriesnot in index	1484#38	Arena (LMArena)

Multimodal

Benchmark	max	not stated	Source	Trend
LiveBench Designnot in indexeffort not stated	–	77.3%#21	LiveBench
AA-3Dnot in index	787#19	–	Artificial Analysis

Long context

Benchmark	low	max	Source	Trend
AA-LCR	72.7%#184	79.7%#87	Artificial Analysis
MLCRnot in index	–	48.3%#13	Artificial Analysis

Composite

Benchmark	low	max	not stated	Source
Epoch Capabilities Indexnot in indexeffort not stated	–	–	156.0#31	Epoch AI Benchmarking Hub
LiveBench 2.0effort not stated	–	–	60.7%#38	LiveBench
LiveBench 1.0not in indexeffort not stated	–	–	76.1%#32	LiveBench
AA Intelligence Index v4.3.2	34.3#95	44.8#39	–	Artificial Analysis
Vals Indexnot in index	–	59.7#9	–	Vals AI
AA Cyber Indexnot in index	–	36.4#10	–	Artificial Analysis

Safety & honesty

Benchmark	max	Source	Trend
FORTRESSnot in index	28.2%#26	Scale AI SEAL

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

#4τ³-Banking (AA)50.3% · of 202

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
Baseten (fp8)	152 tok/s	0.76 s	$2.10	$6.60	1.0M	fp8
Baseten (fp8)	144 tok/s	0.82 s	$2.10	$6.60	1.0M	fp8
Mistral	142 tok/s	0.95 s	$1.54	$4.84	1.0M	nvfp4
ModelRun [by Modular]	126 tok/s	0.60 s	$2.10	$6.60	1.0M	fp4
Fireworks	118 tok/s	1.31 s	$2.10	$6.60	1.0M	–
inference.net	117 tok/s	0.76 s	$0.080	$4.38	1.0M	fp4
Mistral	113 tok/s	1.28 s	$1.40	$4.40	1.0M	nvfp4
inference.net	113 tok/s	0.81 s	$0.080	$4.38	1.0M	fp4
Wafer	111 tok/s	0.43 s	$0.079	$1.40	1.0M	fp4
Mistral	109 tok/s	1.12 s	$1.40	$4.40	1.0M	nvfp4
Crusoe	109 tok/s	0.46 s	$1.40	$4.40	1.0M	fp4
Wafer	107 tok/s	0.40 s	$0.179	$1.40	1.0M	fp4
Decart	106 tok/s	0.95 s	$0.840	$2.69	1.0M	fp4
Modal	100 tok/s	0.74 s	$1.40	$4.40	1.0M	–
Prime Intellect	87 tok/s	1.52 s	$1.40	$4.40	1.0M	–
DigitalOcean	87 tok/s	0.68 s	$0.910	$2.86	1.0M	–
Makora	86 tok/s	1.60 s	$0.140	$4.40	1.0M	fp4
Baseten (fp4)	84 tok/s	1.56 s	$1.40	$4.40	1.0M	fp4
Baseten (fp4)	78 tok/s	2.03 s	$1.40	$4.40	1.0M	fp4
Fireworks	78 tok/s	3.82 s	$0.980	$3.08	1.0M	–
CoreWeave	77 tok/s	0.88 s	$1.20	$4.00	1.0M	fp8
AkashML	74 tok/s	1.68 s	$0.190	$4.40	1.0M	fp8
Alibaba Cloud Int.	72 tok/s	1.12 s	$2.80	$8.80	1M	–
Friendli	70 tok/s	0.46 s	$1.26	$3.96	1.0M	–
Parasail	69 tok/s	0.90 s	$1.40	$4.40	1.0M	fp8
Venice	69 tok/s	2.24 s	$1.40	$4.40	1M	–
Fireworks	68 tok/s	1.17 s	$1.40	$4.40	1.0M	–
Alibaba Cloud Int.	68 tok/s	1.60 s	$1.19	$3.74	1M	–
Relace	65 tok/s	2.06 s	$0.750	$2.40	1.0M	–
Inceptron	64 tok/s	0.88 s	$0.550	$1.00	1.0M	fp4
NovitaAI	60 tok/s	1.57 s	$0.630	$1.98	1.0M	fp8
Z.ai	59 tok/s	2.54 s	$1.40	$4.40	1.0M	fp8
Nebius Token Factory	55 tok/s	0.86 s	$1.40	$4.40	1.0M	fp4
Sail Research	53 tok/s	1.44 s	$0.750	$1.00	1.0M	fp8
Phala	51 tok/s	2.58 s	$0.840	$2.64	1.0M	–
io.net	50 tok/s	1.62 s	$0.490	$3.20	262k	fp8
Together	49 tok/s	4.57 s	$1.40	$4.40	1.0M	–
Baidu Qianfan	47 tok/s	1.65 s	$1.40	$4.40	1.0M	fp8
Morph	45 tok/s	0.90 s	$0.275	$4.40	1.0M	fp8
Sail Research	45 tok/s	1.51 s	$0.750	$1.00	1.0M	fp8
DekaLLM	45 tok/s	2.52 s	$0.180	$2.95	1.0M	–
AtlasCloud	44 tok/s	2.54 s	$1.40	$4.40	1.0M	fp8
Reka AI	40 tok/s	1.20 s	$0.140	$4.20	1.0M	–
Tenstorrent	38 tok/s	1.90 s	$0.600	$2.50	1.0M	fp8
DeepInfra	35 tok/s	29 s	$0.563	$2.50	1.0M	fp4
SiliconFlow	33 tok/s	4.68 s	$0.700	$2.20	1.0M	fp8
Cloudflare	28 tok/s	1.82 s	$1.40	$4.40	1.0M	–
GMICloud	27 tok/s	3.63 s	$0.700	$2.20	1.0M	fp8

Price history

Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.

input outputnow $0.087 in · $6.00 out

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.260 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0019	$0.0015	29.9 s
Summarize a 30-page report	12,000 / 600	$0.019	$0.0092	33.5 s
Code edit	6,000 / 1,500	$0.015	$0.0099	44.2 s
Agentic coding session	60,000 / 4,000	$0.102	$0.050	1.2 min
Structured extraction	2,000 / 200	$0.0037	$0.0020	28.7 s