Zhipu AI Open weights: zai-org/GLM-5.2Reasoning modelCompare with another model

GLM 5.2

Reasoning effort

GLM 5.2 is a Zhipu AI open-weights reasoning model, released 16 Jun 2026. Its best configuration (max reasoning effort) ranks #85 of 775 on the BenchLeader Index at 62.4 ±3.6, in the upper half. The ± is the point: 168 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 instruction following (72) and lowest in knowledge (54). At $2.15 per million tokens blended it is pricier than most ranked models. Output speed of 88 tokens per second puts it faster than most, with a first answer in 26.2 s. It has been measured at 9 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: 88 tok/s; measured by Artificial Analysis
First answer: 26 s; first token 0.49 s
Context: 1M
Full answer: 32 s; median, reasoning included
Cost per run: $1.47; one full Intelligence Index run
Released: 16 Jun 2026

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

Overall index62
Reasoning70
Coding62
Agents & tools63
Maths57
Knowledge54
Instruction following72
Human preference67
Long context64
Composite67

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 5.3max	18 Aug 2026	63.6	#59
GLM 5.2maxthis page	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
no reasoning	48.0	#434	180 tok/s	1.11 s	$0.0019	Agents & tools 45 · Coding 35 · Composite 55 · Knowledge 59 · Long context 46 · Maths 45 · Reasoning 49
minimal	–	–	127 tok/s	0.49 s	$0.0019
low	–	–	127 tok/s	0.49 s	$0.0019	Maths 56 · Reasoning 63
medium	–	–	127 tok/s	0.49 s	$0.0019
high	–	–	127 tok/s	0.49 s	$0.0019	Coding 62
xhigh	–	–	127 tok/s	0.49 s	$0.0019
maxbest	62.4	#85	88 tok/s	26 s	$0.0019	Agents & tools 63 · Coding 62 · Composite 67 · Human preference 67 · Instruction following 72 · Knowledge 54 · Long context 64 · Maths 57 · Reasoning 70
thinking	–	–	127 tok/s	0.49 s	$0.0019	Reasoning 55
not stated	52.0	#326	127 tok/s	0.49 s	$0.0019	Agents & tools 52 · Coding 49 · Human preference 54 · Knowledge 59 · Maths 47 · Reasoning 53

How its index has moved

8 Sept 2026 to 20 Oct 2026

max
not stated
no reasoning

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.

Reasoning

Benchmark	no reasoning	minimal	low	medium	max	thinking	not stated	Source
GPQA Diamond	71.2%#167	–	87.9%#64	–	91.9%#28	–	–	Epoch AI Benchmarking Hub
SimpleBencheffort not stated	–	–	–	–	–	–	58.8%#32	SimpleBench
Arena Hard Prompts	–	–	–	–	1496#39	–	–	Arena (LMArena)
LiveBench 1.0 Reasoningnot in indexeffort not stated	–	–	–	–	–	–	78.6%#55	LiveBench
GPQA Diamond (AA)not in index	68.6%#308	–	–	–	89.5%#74	–	–	Artificial Analysis
Humanity's Last Exam (AA)not in index	9.8%#325	–	–	–	41.1%#78	–	–	Artificial Analysis
GPQA Diamond (Vals)not in indexeffort not stated	–	–	–	–	–	–	85.6%#44	Vals AI
Kagi LLM Benchmark	–	–	–	–	–	60.0%#50	62.6%#44	Kagi LLM Benchmark
ARC-AGI-1effort not stated	–	–	–	–	–	–	77.0%#118	ARC Prize
ARC-AGI-2effort not stated	–	–	–	–	–	–	22.8%#131	ARC Prize
CritPt	3.1%#173	–	–	–	20.9%#56	–	–	Artificial Analysis
Chess Puzzlesnot in index	19.0%#94	–	18.0%#98	–	22.0%#83	–	–	Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index	18.0%#68	18.0%#68	19.0%#63	15.0%#90	–	–	–	Epoch AI Benchmarking Hub
LMCAnot in index	–	–	–	–	45.8%#86	–	–	Epoch AI Benchmarking Hub
DTBenchnot in index	–	–	–	–	93.6%#63	–	–	Epoch AI Benchmarking Hub

Coding

Benchmark	no reasoning	high	max	not stated	Source
SWE-bench Verified (Epoch)	–	–	78.7%#5	–	Epoch AI Benchmarking Hub
SciCode	36.1%#185	–	51.2%#99	–	SciCode
WeirdML	–	67.3%#36	70.1%#33	–	WeirdML
FrontierCode	24.5%#37	–	–	–	Cognition
Arena Coding	–	–	1514#48	–	Arena (LMArena)
Arena WebDev	–	–	1604#27	–	Arena (LMArena)
LiveBench 1.0 Codingnot in indexeffort not stated	–	–	–	79.7%#23	LiveBench
SciCode (AA)not in index	–	–	51.2%#98	–	Artificial Analysis
LiveCodeBencheffort not stated	–	–	–	69.5%#89	Vals AI
SWE-bench (Vals)not in index	–	–	82.8%#18	–	Vals AI
Code Migrationnot in index	–	–	37.9%#33	–	Vals AI
ProgramBenchnot in index	–	–	0.5%#18	–	Vals AI
Vibe Code Bench v1.1not in index	–	–	64.0%#50	–	Vals AI
SWE Atlas: Codebase QnAnot in indexeffort not stated	–	–	–	48.1%#5	Scale AI SEAL
SWE Atlas: Refactoringnot in indexeffort not stated	–	–	–	42.4%#9	Scale AI SEAL
SWE Atlas: Test Writingnot in indexeffort not stated	–	–	–	41.5%#10	Scale AI SEAL
Surface Evolver Benchnot in index	–	55.6%#15	–	–	Epoch AI Benchmarking Hub
DeepSWE v1.1not in index	–	36.3%#59	43.8%#54	–	Epoch AI Benchmarking Hub
ALE-Benchnot in index	–	1047.0#54	1010.2#56	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	–	–	1.0%#125	–	Artificial Analysis
Terminal-Bench 2.1 (AA)not in index	51.7%#113	–	77.9%#54	–	Artificial Analysis
Startrise Frontendnot in indexeffort not stated	–	–	–	72.1%#14	Startrise

Agents & tools

Benchmark	no reasoning	xhigh	max	not stated	Source
APEX-Agentseffort not stated	–	–	–	45.2%#37	Mercor
Arena Agent	–	–	2.1#21	–	Arena (LMArena)
LiveBench Agentic Codingnot in indexeffort not stated	–	–	–	51.8%#39	LiveBench
Terminal-Bench Hard	–	–	50.8%#22	–	Artificial Analysis
τ²-Bench Telecom (AA)not in index	–	–	99.1%#1	–	Artificial Analysis
Terminal-Bench 2.1 (Vals)	–	–	67.8%#32	–	Vals AI
MCP Atlaseffort not stated	–	–	–	77.8%#15	Scale AI SEAL
HiL-Bencheffort not stated	–	–	–	43.7%#13	Scale AI SEAL
τ³-bench Bankingnot in index	–	37.1%#11	–	–	τ²-bench
GDPval-AA v2.1not in index	37.4%#110	–	43.6%#82	–	Artificial Analysis
τ³-Banking (AA)	16.7%#107	–	34.6%#50	–	Artificial Analysis
ITBench SRE (AA)not in index	–	–	42.7%#14	–	Artificial Analysis
APEX-Agents (AA)not in index	–	–	33.7%#8	–	Artificial Analysis
Finance Agent v2not in indexeffort not stated	–	–	–	49.7%#45	Vals AI
Legal Research Benchnot in index	–	–	31.3%#45	–	Vals AI
SkillsBenchnot in index	–	–	45.1%#26	–	Vals AI
SREBenchnot in index	–	–	0.0%#31	–	Vals AI
PostTrainBenchnot in index	–	–	31.7%#7	–	Epoch AI Benchmarking Hub
GBAEvalnot in indexeffort not stated	–	–	–	0.0%#21	Epoch AI Benchmarking Hub
Vending-Bench 2not in indexeffort not stated	–	–	–	8313.8#12	Andon Labs
AutomationBenchnot in index	–	–	28.4%#118	–	Artificial Analysis
GDP.pdf (AA)not in index	–	–	10.4%#125	–	Artificial Analysis
EnterpriseOps-Gymnot in index	–	–	42.7%#21	–	Artificial Analysis

Maths

Benchmark	no reasoning	low	max	not stated	Source
FrontierMath Tiers 1–3	42.5%#59	54.7%#49	59.2%#40	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	–	–	29.3%#39	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	28.9%#208	75.6%#125	86.4%#92	–	Epoch AI Benchmarking Hub
ProofBench	–	–	35.0%#41	–	Vals AI
LiveBench Mathematicsnot in indexeffort not stated	–	–	–	89.8%#37	LiveBench
AIME 2026effort not statedpossible contamination	–	–	–	90.0%#27	MathArena
HMMT February 2026effort not statedpossible contamination	–	–	–	92.4%#13	MathArena
MathArena Apexeffort not statedpossible contamination	–	–	–	19.8%#15	MathArena
Arena Maths	–	–	1485#33	–	Arena (LMArena)

Knowledge

Benchmark	no reasoning	xhigh	max	not stated	Source
SimpleQA Verified	–	–	34.2%#54	–	Epoch AI Benchmarking Hub
LiveBench 1.0 Data Analysisnot in indexeffort not stated	–	–	–	73.7%#41	LiveBench
AA-Omniscience	-6.6#168	–	4.4#111	–	Artificial Analysis
MMLU-Proeffort not stated	–	–	–	86.7%#38	Vals AI
LegalBencheffort not stated	–	–	–	84.1%#40	Vals AI
CorpFineffort not stated	–	–	–	66.1%#21	Vals AI
TaxEvaleffort not stated	–	–	–	73.3%#48	Vals AI
Excel Modeling Benchmarknot in indexeffort not stated	–	–	–	61.5%#33	Vals AI
MedCodenot in indexeffort not stated	–	–	–	40.8%#58	Vals AI
MedScribenot in indexeffort not stated	–	–	–	83.5%#44	Vals AI
DrugDiscoveryBenchnot in index	–	36.2%#11	–	–	Scale AI SEAL
EBR-benchnot in index	–	–	9.5%#22	–	Epoch AI Benchmarking Hub
AA-Omniscience: accuracynot in index	20.3%#299	–	24.3%#241	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	66.3%#60	–	73.7%#30	–	Artificial Analysis

Instruction following

Benchmark	max	not stated	Source
LiveBench Languagenot in indexeffort not stated	–	76.2%#45	LiveBench
LiveBench Instruction Followingnot in indexeffort not stated	–	62.3%#54	LiveBench
IFBench	73.3%#41	–	Artificial Analysis
Arena Instruction Followingnot in index	1470#38	–	Arena (LMArena)

Human preference

Benchmark	max	not stated	Source
Arena Text	1476#32	–	Arena (LMArena)
EQ-Bench 4effort not stated	–	1222#13	EQ-Bench
Arena Creative Writingnot in index	1455#32	–	Arena (LMArena)
Arena Multi-turnnot in index	1476#48	–	Arena (LMArena)
Arena Longer Queriesnot in index	1487#33	–	Arena (LMArena)

Long context

Benchmark	no reasoning	max	Source	Trend
AA-LCR	42.3%#345	78.3%#111	Artificial Analysis
MLCRnot in index	–	7.2%#69	Artificial Analysis

Composite

Benchmark	no reasoning	max	not stated	Source
Epoch Capabilities Indexnot in indexeffort not stated	–	–	151.9#51	Epoch AI Benchmarking Hub
LiveBench 1.0not in indexeffort not stated	–	–	73.2%#46	LiveBench
AA Intelligence Index v4.3.2	22.4#198	33.7#102	–	Artificial Analysis

Where it wins

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

#5SWE-bench Verified (Epoch)78.7% · of 33

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
Decart	294 tok/s	0.67 s	$1.20	$8.00	1.0M	fp4
Baseten	190 tok/s	0.52 s	$2.10	$6.60	1.0M	fp8
Baseten	180 tok/s	0.57 s	$2.10	$6.60	1.0M	fp8
CoreWeave	158 tok/s	0.56 s	$0.620	$2.42	1.0M	fp4
Nebius Token Factory	143 tok/s	0.93 s	$1.40	$4.40	1.0M	fp4
Wafer	139 tok/s	0.38 s	$0.034	$7.00	1.0M	fp4
Parasail	124 tok/s	0.71 s	$1.40	$4.40	262k	fp4
Together	120 tok/s	0.42 s	$1.40	$4.40	1.0M	–
Wafer	117 tok/s	0.40 s	$0.190	$8.00	1.0M	fp4
Inceptron	103 tok/s	0.37 s	$0.690	$3.15	1.0M	fp4
Baseten	102 tok/s	1.68 s	$1.40	$4.40	1.0M	fp8
Alibaba Cloud Int.	99 tok/s	1.23 s	$0.966	$3.04	1M	fp8
Mistral (nvfp4)	97 tok/s	0.89 s	$1.40	$4.40	1.0M	nvfp4
Friendli	88 tok/s	0.35 s	$1.40	$4.40	1.0M	–
Alibaba Cloud Int.	86 tok/s	1.52 s	$2.31	$7.26	1M	fp8
DigitalOcean	86 tok/s	0.59 s	$0.700	$2.20	1.0M	–
Baseten	81 tok/s	1.83 s	$1.40	$4.40	1.0M	fp8
Cloudflare	80 tok/s	0.90 s	$0.190	$8.00	262k	–
SiliconFlow	74 tok/s	1.56 s	$0.700	$2.20	1.0M	fp8
Venice	67 tok/s	1.59 s	$1.40	$4.40	1M	fp8
Baidu Qianfan (fp4)	66 tok/s	1.02 s	$2.25	$7.88	1.0M	fp4
Alibaba Cloud Int.	61 tok/s	1.89 s	$1.12	$3.52	1M	fp8
Baidu Qianfan (fp8)	59 tok/s	1.05 s	$1.40	$4.40	1.0M	fp8
Z.ai	58 tok/s	2.54 s	$1.40	$4.40	1.0M	fp8
StreamLake	55 tok/s	1.23 s	$0.613	$1.93	1.0M	fp8
DeepInfra	53 tok/s	1.53 s	$0.563	$1.80	1.0M	fp4
Morph	50 tok/s	0.69 s	$0.275	$4.40	1.0M	fp8
NovitaAI	49 tok/s	2.07 s	$0.650	$2.04	1.0M	fp8
AtlasCloud	46 tok/s	1.31 s	$0.938	$2.95	1.0M	fp8
GMICloud	44 tok/s	7.94 s	$1.40	$4.40	1.0M	fp8
DeepInfra	35 tok/s	4.09 s	$0.450	$1.44	1.0M	fp4
Phala	18 tok/s	2.94 s	$1.26	$3.00	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 $0.556 in · $1.75 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	29.6 s
Summarize a 30-page report	12,000 / 600	$0.019	$0.0092	33.0 s
Code edit	6,000 / 1,500	$0.015	$0.0099	43.3 s
Agentic coding session	60,000 / 4,000	$0.102	$0.050	1.2 min
Structured extraction	2,000 / 200	$0.0037	$0.0020	28.5 s