Zhipu AI Open weights: zai-org/GLM-5.3-FlashReasoning modelAuto-detected

GLM-5.3-Flash

Name: GLM-5.3-Flash benchmark results
Creator: BenchLeader

Best configuration ranks #74 of 610 on the BenchLeader Index at 61.0 ±6.2. Last measured 8 Sept 2026. Released 20 Aug 2026.

Blended price: $0.119/M; $0.075 in · $0.250 out
Output speed: 72 tok/s
First answer: 30 s; first token 1.70 s
Context: 1M

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

Overall index61
Reasoning67
Coding64
Agents & tools52
Knowledge68
Human preference67
Multimodal65
Long context67
Composite61

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. “Default” means the publisher did not say which setting was used.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
max	55.6	#173	72 tok/s	30 s	$0.0001	Agents & tools 52 · Coding 58 · Knowledge 59 · Maths 53 · Reasoning 65
defaultbest	61.0	#74	72 tok/s	30 s	$0.0001	Agents & tools 52 · Coding 64 · Composite 61 · Human preference 67 · Knowledge 68 · Long context 67 · Multimodal 65 · Reasoning 67

Benchmark results

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	max	default	Source
GPQA Diamond	90.2%#37	–	Epoch AI Benchmarking Hub
LMArena Hard Prompts	–	1496#31	LMArena
LiveBench Reasoningnot in index	–	77.6%#43	LiveBench
GPQA Diamond (AA)not in index	–	91.2%#49	Artificial Analysis
Humanity's Last Exam (AA)not in index	–	39.9%#56	Artificial Analysis
GPQA Diamond (Vals)not in index	86.4%#41	–	Vals AI

Coding

Benchmark	max	default	Source
SciCode	–	46.1%#81	SciCode
LMArena Coding	–	1534#9	LMArena
LMArena WebDev	–	1605#14	LMArena
LiveBench Codingnot in index	–	79.0%#17	LiveBench
SciCode (AA)not in index	–	51.6%#56	Artificial Analysis
LiveCodeBench	80.5%#66	–	Vals AI
SWE-bench (Vals)not in index	92.0%#10	–	Vals AI

Agents & tools

Benchmark	max	default	Source
LMArena Agent	–	2#20	LMArena
LiveBench Agentic Codingnot in index	–	56.8%#19	LiveBench
Terminal-Bench 2.1 (Vals)	62.9%#30	–	Vals AI

Maths

Benchmark	max	default	Source
FrontierMath Tiers 1–3	55.8%#33	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	17.1%#42	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	93.9%#42	–	Epoch AI Benchmarking Hub
ProofBench	21.0%#34	–	Vals AI
LiveBench Mathematicsnot in index	–	81.2%#45	LiveBench

Knowledge

Benchmark	max	default	Source
LiveBench Data Analysisnot in index	–	76.4%#24	LiveBench
AA-Omniscience	–	7.5#66	Artificial Analysis
MMLU-Pro	86.1%#47	–	Vals AI
LegalBench	83.9%#41	–	Vals AI
TaxEval	75.6%#16	–	Vals AI

Instruction following

Benchmark	max	default	Source	Trend
LiveBench Languagenot in index	–	77.3%#32	LiveBench

Human preference

Benchmark	max	default	Source	Trend
LMArena Text	–	1474#29	LMArena

Multimodal

Benchmark	max	default	Source	Trend
LMArena Vision	–	1296#15	LMArena

Long context

Benchmark	max	default	Source	Trend
AA-LCR	–	80.0%#51	Artificial Analysis

Composite

Benchmark	max	default	Source
Epoch Capabilities Indexnot in index	–	151.4#39	Epoch AI Benchmarking Hub
LiveBench	–	71.6%#40	LiveBench
AA Intelligence Index	–	41.9#30	Artificial Analysis
Vals Indexnot in index	47.2#30	–	Vals AI

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
Morph	135 tok/s	0.99 s	$0.388	$1.36	1.0M	fp8
Crusoe	115 tok/s	0.95 s	$0.150	$0.500	1.0M	fp4
Baseten	102 tok/s	0.80 s	$0.150	$0.500	1.0M	fp8
Friendli	82 tok/s	1.03 s	$0.150	$0.500	1.0M	–
Modal	71 tok/s	0.75 s	$0.150	$0.500	1.0M	fp8
Relace	70 tok/s	0.95 s	$0.120	$0.400	1.0M	fp4
CoreWeave	65 tok/s	1.11 s	$0.150	$0.500	1.0M	fp8
Together	62 tok/s	1.09 s	$0.150	$0.500	1.0M	–
Parasail	57 tok/s	1.22 s	$0.150	$0.500	1.0M	fp8
Makora	55 tok/s	1.04 s	$0.140	$0.470	1.0M	–
Reka AI	54 tok/s	1.72 s	$0.150	$0.500	262k	fp8
Fireworks	49 tok/s	1.39 s	$0.150	$0.500	1.0M	–
Cloudflare	44 tok/s	1.68 s	$0.150	$0.500	1.3M	–
Z.ai	39 tok/s	5.10 s	$0.075	$0.250	1.0M	fp8
NovitaAI	38 tok/s	3.37 s	$0.132	$0.440	1.0M	fp8
Phala	34 tok/s	3.43 s	$0.150	$0.500	1.0M	fp8
SiliconFlow	34 tok/s	2.35 s	$0.150	$0.500	1.0M	fp8
StreamLake	32 tok/s	4.31 s	$0.135	$0.450	1.0M	fp8
NextBit	31 tok/s	1.78 s	$0.150	$0.500	1.0M	fp8
Sail Research	27 tok/s	1.38 s	$0.150	$0.500	1.0M	fp8
io.net	25 tok/s	1.77 s	$0.150	$0.500	262k	fp8
DigitalOcean	19 tok/s	2.78 s	$0.150	$0.500	1.0M	–
GMICloud	18 tok/s	3.35 s	$0.113	$0.375	1.0M	fp8
Venice	13 tok/s	3.92 s	$0.150	$0.500	1.0M	–
DeepInfra	12 tok/s	2.34 s	$0.075	$0.250	1.0M	fp4
Wafer	10 tok/s	2.13 s	$0.100	$0.350	1.0M	–

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 $0.015 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0001	$0.0001	34.5 s
Summarise a 30-page report	12,000 / 600	$0.0011	$0.0005	38.7 s
Code edit	6,000 / 1,500	$0.0008	$0.0006	51.2 s
Agentic coding session	60,000 / 4,000	$0.0055	$0.0028	1.4 min
Structured extraction	2,000 / 200	$0.0002	$0.0001	33.1 s