Zhipu AI Open weights: zai-org/GLM-5Reasoning model

GLM-5

Name: GLM-5 benchmark results
Creator: BenchLeader

Best configuration ranks #112 of 610 on the BenchLeader Index at 59.0 ±4.0. Last measured 8 Sept 2026. Released 11 Feb 2026.

Blended price: $1.55/M; $1.00 in · $3.20 out
Output speed: 66 tok/s
First answer: 48 s; first token 1.35 s
Context: 205k

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

Overall index59
Reasoning51
Coding53
Agents & tools57
Maths53
Knowledge64
Instruction following70
Human preference65
Long context64
Composite64

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
no reasoning	54.5	#197	54 tok/s	1.51 s	$0.0014	Agents & tools 68 · Composite 56 · Instruction following 56 · Knowledge 58 · Long context 48
high	–	–	66 tok/s	48 s	$0.0010	Coding 64
thinking	57.0	#150	66 tok/s	48 s	$0.0010	Agents & tools 51 · Coding 56 · Knowledge 57 · Maths 63 · Reasoning 68
defaultbest	59.0	#112	66 tok/s	48 s	$0.0014	Agents & tools 57 · Coding 53 · Composite 64 · Human preference 65 · Instruction following 70 · Knowledge 64 · Long context 64 · Maths 53 · Reasoning 51

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	no reasoning	high	thinking	default	Source
GPQA Diamond	–	–	–	87.8%#58	Epoch AI Benchmarking Hub
SimpleBench	–	–	–	53.2%#41	SimpleBench
LMArena Hard Prompts	–	–	–	1478#56	LMArena
GPQA Diamond (AA)not in index	66.6%#315	–	–	82.0%#170	Artificial Analysis
Humanity's Last Exam (AA)not in index	7.7%#287	–	–	29.3%#109	Artificial Analysis
GPQA Diamond (Vals)not in index	–	–	83.3%#56	–	Vals AI
Kagi LLM Benchmark	–	–	75.0%#12	51.7%#80	Kagi LLM Benchmark
ARC-AGI-1	–	–	–	44.7%#126	ARC Prize
ARC-AGI-2	–	–	–	4.9%#120	ARC Prize

Coding

Benchmark	no reasoning	high	thinking	default	Source
SWE-bench Verified (Epoch)	–	–	–	72.1%#22	Epoch AI Benchmarking Hub
WeirdML	–	–	–	48.2%#70	WeirdML
LMArena Coding	–	–	–	1498#67	LMArena
LMArena WebDev	–	–	–	1436#60	LMArena
LiveCodeBench	–	–	81.9%#57	–	Vals AI
IOI	–	–	22.0%#20	–	Vals AI
SWE-bench (Vals)not in index	–	–	71.4%#55	–	Vals AI
SWE-bench Verified (bash only)	–	72.8%#7	–	–	SWE-bench
SWE-bench Verified (any scaffold)not in index	–	72.8%#11	–	–	SWE-bench

Agents & tools

Benchmark	no reasoning	high	thinking	default	Source
Terminal-Bench	–	–	–	52.4%#23	Terminal-Bench
APEX-Agents	–	–	–	17.2%#43	Mercor
Terminal-Bench Hard	39.4%#56	–	–	43.2%#39	Artificial Analysis
τ²-Bench Telecom (AA)not in index	97.4%#12	–	–	98.3%#8	Artificial Analysis
τ²-bench	–	–	81.0%#6	–	τ²-bench

Maths

Benchmark	no reasoning	high	thinking	default	Source
OTIS Mock AIME	–	–	–	80.0%#99	Epoch AI Benchmarking Hub
AIME (Vals)	–	–	91.7%#23	–	Vals AI
AIME 2026	–	–	–	96.7%#6	MathArena
HMMT February 2026	–	–	–	86.4%#20	MathArena
MathArena Apex	–	–	–	10.9%#21	MathArena

Knowledge

Benchmark	no reasoning	high	thinking	default	Source
AA-Omniscience	-11.8#152	–	–	0.3#92	Artificial Analysis
MMLU-Pro	–	–	86.0%#49	–	Vals AI
LegalBench	–	–	84.1%#38	–	Vals AI
CorpFin	–	–	62.9%#47	–	Vals AI
TaxEval	–	–	70.0%#92	–	Vals AI
MedQA	–	–	94.3%#18	–	Vals AI

Instruction following

Benchmark	no reasoning	high	thinking	default	Source
IFBench	55.2%#144	–	–	72.3%#49	Artificial Analysis

Human preference

Benchmark	no reasoning	high	thinking	default	Source
LMArena Text	–	–	–	1458#54	LMArena

Long context

Benchmark	no reasoning	high	thinking	default	Source
AA-LCR	43.7%#274	–	–	75.7%#107	Artificial Analysis

Composite

Benchmark	no reasoning	high	thinking	default	Source
Epoch Capabilities Indexnot in index	–	–	–	145.9#64	Epoch AI Benchmarking Hub
AA Intelligence Index	21.8#157	–	–	27.9#95	Artificial Analysis

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
Amazon Bedrock	67 tok/s	1.48 s	$1.00	$3.20	203k	–
Baidu Qianfan	58 tok/s	0.94 s	$0.700	$2.24	203k	fp8
StreamLake	49 tok/s	2.41 s	$0.600	$1.92	198k	fp8
NovitaAI	49 tok/s	1.19 s	$1.00	$3.20	203k	fp8
GMICloud	48 tok/s	1.35 s	$0.600	$1.92	203k	fp8
Z.ai	48 tok/s	5.79 s	$1.00	$3.20	203k	fp8
DeepInfra	43 tok/s	0.89 s	$0.600	$2.08	203k	fp4
Venice	42 tok/s	1.24 s	$1.00	$3.20	198k	fp8
SiliconFlow	36 tok/s	2.40 s	$0.950	$2.55	205k	fp8

Price history

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

input outputnow $0.600 in · $1.92 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 $0.200 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0014	$0.0011	52.8 s
Summarise a 30-page report	12,000 / 600	$0.014	$0.0067	57.4 s
Code edit	6,000 / 1,500	$0.011	$0.0072	1.2 min
Agentic coding session	60,000 / 4,000	$0.073	$0.037	1.8 min
Structured extraction	2,000 / 200	$0.0026	$0.0014	51.3 s