GLM 5.3

Aug 2026

GLM 5.3

Compare in

on

$2.01per task?

$1.40 / $4.40per 1M in / out?cache $0.26

186 tok/sfirst token in 1.2 s?

1Mcontext

Rankingat max effort?

Strong at

S

Coding#12 of 132 Agents & work tasks#12 of 175 Science#23 of 312

A

Reasoning#37 of 334 Long context#48 of 273 Knowledge#32 of 318

Alternatives?

ModelScoreCostPer taskWhat changes

CheaperGPT-6.1 Solmax×3.4-64%$0.72Remove unsupported parameters+5

BetterGPT-6 Lunamax+92%-97%$0.068Function calling only with none · Chat Completions+3

OpenGLM 5.2maxsame-27%$1.47

CheaperGPT-6 Astrahigh×3.3-14%$1.73Monitoring can return HTTP 403+2

CheaperGPT-6 Solmax×3.1-48%$1.04

Run it yourself

estimate

Fits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU

Licenceglm-5.3commercial use with conditions

History

10 versions

Benchmarks

all categories

Chat & Knowledge

Humanity's Last Exam42.3#37

Arena Text · overall1,479#21

Arena Text · Chinese1,528.8#14

Arena Text · English1,480#17

Arena Text · French1,501.6#10

Arena Text · German1,492.9#9

Arena Text · instruction following1,476.3#18

Arena Text · Japanese1,454.7#17

Arena Text · Korean1,466.3#6

Arena Text · non-English1,457.7#24

Arena Text · Polish1,468.4#16

Arena Text · Russian1,464.1#26

Arena Text · Spanish1,464.4#25

AA Omniscience Index14.3#32

SimpleQA Verified v141#39

CritPt (physics)19.1#32

AA Intelligence Index 4.3.244.8#18

AA Omniscience · Accuracy33.9#63

AA Omniscience · Hallucination Rate29.6#32

AA-LCR (long context) 1.179.7#48

GDP.pdf · AA all-pass11.2#52

FrontierMath Tiers 1–3 (v2)68.8#25

FrontierMath Tier 4 (v2)29.3#29

Chess Puzzles37#31

Mystery Game Puzzles33#21

HLE text-onlylab42

ExploitGymlab15

AA Cyber Index successeslab36

Arena Text · overall · full cohort1,471.5#19

GPQA Diamond91.7#23

GPQA Diamond · Epoch90.9#24

OTIS Mock AIME 2024 202591.1#43

Coding

Terminal-Bench 4.041.9#12

DeepSWE 1.169#11

FrontierCode 1.140.1#16

WebDev Arena1,621#21

SWE-bench Pro (v2 full)95.6#9

SWE-bench Pro (v2 hard)84.3#9

Terminal-Bench Science 0.19.5#15

SciCode Verified pass@1 n=6lab92.5

Terminal-Bench 3.0lab28.3

ProgramBenchlab19

NL2Repo-Benchlab58

SWE Atlas QnAlab59

Surge Human Eval on Codelab3.6

WebDev Arena · full cohort1,621.2#21

Terminal-Bench 2.183.9#16

Agents & Tool Use

AutomationBench-AA62.2#13

GDPval-AA (work tasks) 2.157.5#12

τ³-Banking · AA50.3#3

AA Analyst Agent38.8#21

ITBench SRE46.1#11

EnterpriseOps-Gym-AA36.4#27

AA-Briefcase 1.1 · Elo1,506.9#18

Harvey LAB-AA 1.10.3#17

AutomationBenchlab62.2

CyberGymlab84.5

HLE with toolslab62.5

Automation-Benchlab48.8

Agents Last Examlab28.5

Finance Agent v2lab55.8

Harvey Legal Agent Benchmarklab8.3

CyberGym-E2E AAlab29

Cybenchlab85

Finch FinWorkBenchlab65.1

B3 Attack Resistancelab93.3

Harmful cyber request refusal meanlab87

AA Briefcase combined Elolab1,510

KORA aggregatelab1.7

Writing & Language

Arena Text · creative writing1,453#23

Lech Mazur Short-Story3#5

Arena Text · creative writing · full cohort1,456#18

Watch out 4

Reasoning cannot be disabled›

Disabling reasoning is unsupported; keep reasoning enabled. Source ↗

tool_choice auto · Z.ai API›

The first-party chat reference exposes only auto for tool_choice. Source ↗

Complete unmodified reasoning_content›

Preserved thinking defaults on for Coding Plan and off for the standard API; when enabled, return complete unmodified reasoning_content in its original order to protect quality and cache hits. Source ↗

Only one stop word›

Only one stop word is supported, so do not rely on a list of alternative stop sequences. Source ↗

FAQ

How much does GLM 5.3 cost?›

$1.40 per 1M input tokens and $4.40 per 1M output tokens; cached input $0.26 per 1M; cheapest via Wafer at $0.079 / $1.40. A typical benchmark task costs about $2.01 at max effort.

Is GLM 5.3 free to use?›

There is no free API; the cheapest paid price is $0.079 per 1M input tokens. The weights are open, so you can run it yourself.

How good is GLM 5.3, and what is it best at?›

#37 of 334 in Chat & Knowledge, #12 of 132 in Coding, #13 of 133 in Agents & Tool Use, #23 of 195 in Writing & Language, #52 of 133 in Vision & Documents. Strongest at: Coding, Agents & work tasks, Reasoning, Science, Long context, Knowledge.

Is GLM 5.3 better than GLM 5.2?›

On FrontierMath Tier 4 (v2), GLM 5.3 scores 29.3 vs 29.3 — about the same. It is 36% more expensive per task.

How can I use GLM 5.3?›

Through the API of AkashML, Alibaba, AtlasCloud, Baidu, Cloudflare and 24 more.

What is the context window of GLM 5.3?›

1M tokens, with up to 131K output tokens.

When was GLM 5.3 released? Is there a newer version?›

Released 2026-08-14. It is the current version in its line.

Is GLM 5.3 open source?›

Yes, the weights are open (glm-5.3 licence).

Can I run GLM 5.3 locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

Can GLM 5.3 read images?›

No, it accepts text only.

What is GLM 5.3?›

GLM 5.3 is an open-weights language model from Z.ai, released Aug 2026, the successor to GLM 5.2.

What are the best alternatives to GLM 5.3?›

GPT-6.1 Sol (cheaper), GPT-6 Luna (better), GLM 5.2 (open weights).