When was GLM-5.3 released?

GLM-5.3 was released by Z.ai on Friday, Aug 14 2026.

GLM-5.3 was built by Z.ai. Chinese AI lab spun out of Tsinghua University (formerly Zhipu AI), building the open-weight GLM family. Rebranded internationally as Z.ai in 2025.

What benchmark scores did GLM-5.3 get?

GLM-5.3 reports 13 tracked benchmark scores — BullshitBench v2: 72%; DeepSWE 1.1: 66.9%; Terminal-Bench 4.0: 41.82%; Terminal-Bench 3.0: 28.3%; Terminal-Bench 2.1: 88.2%; CyberGym: 84.5%; ExploitBench: 54.4%; ExploitGym (6-hour budget): 130; ExploitGym (2-hour budget): 105; Humanity's Last Exam (with tools): 62.5%; Agent's Last Exam (pass@1): 28.5%; AutomationBench: 48.2%; GDPval-AA v2: 1769. Scores are the figures published at release by Z.ai. It holds the best score among all models tracked here on ExploitGym (6-hour budget) and ExploitGym (2-hour budget).

How many parameters does GLM-5.3 have?

GLM-5.3 is reported at 743B parameters.

Is GLM-5.3 open source?

Partly. GLM-5.3 is an open-weight model: the trained weights are free to download and run locally or on your own infrastructure, but the training data and code are not fully released and the license may restrict some commercial uses. It is not open source in the strict sense.

What came before and after GLM-5.3?

Z.ai's previous tracked release was GLM-5.2 on Jun 16 2026, 59 days earlier. It was followed by GLM-5.3-Flash on Aug 26 2026.

GLM-5.3Open Weight

Released Aug 14 2026

GLM-5.3 is an AI model released by Z.ai on Friday, Aug 14 2026, 59 days after GLM-5.2. It is an open-weight model — the trained weights are available to download and run. It is a 743B parameter model. Benchmark results (shown below) cover ExploitGym, BullshitBench v2, DeepSWE 1.1, Terminal-Bench 4.0, Terminal-Bench 3.0, Terminal-Bench 2.1, and 6 more.

Available from


Inceptron	$0.8727	$3.36	1M	fp4
DigitalOcean	$0.95	$3.40	1M	—
Morph	$1.00	$3.41	1M	fp8
Reka	$1.0075	$3.41	256K	fp8
Novita	$1.092	$3.432	1M	fp8
GMICloud	$1.12	$3.52	1M	fp8
Phala	$1.12	$3.52	1M	—
Decart	$1.19	$3.74	1M	fp4
Sail Research	$1.2572	$3.9512	1M	fp8
AkashML	$1.17	$3.96	1M	fp8
Friendli	$1.26	$3.96	1M	—
DeepInfra	$1.20	$4.00	1M	fp4
Wafer	$1.19	$4.40	1M	—
Makora	$1.35	$4.40	980K	fp4
AtlasCloud	$1.40	$4.40	256K	fp8
BaseTen	$1.40	$4.40	1M	fp4
Cloudflare	$1.40	$4.40	1.3M	—
Crusoe	$1.40	$4.40	1M	fp4
Fireworks	$1.40	$4.40	1M	—
Modal	$1.40	$4.40	1M	—
Parasail	$1.40	$4.40	1M	fp8
SiliconFlow	$1.40	$4.40	1M	fp8
Together	$1.40	$4.40	1M	—
Venice	$1.75	$5.50	1M	—
BaseTen	$2.10	$6.60	1M	fp8

USD per 1M tokens

Benchmarks

Coding

DeepSWE 1.1

66.9%

#10 of 26

Terminal & CLI

Terminal-Bench 4.0

41.82%

via Terminal-Bench

#6 of 16

Terminal-Bench 3.0

28.3%

#3 of 7

Terminal-Bench 2.1

88.2%

via BenchLM

#6 of 30

Agentic & tool use

ExploitGym

130

6-hour budget

105

2-hour budget

CyberGym

84.5%

#4 of 10

ExploitBench

54.4%

#2 of 2

Agent's Last Exam

28.5%

pass@1

#3 of 7

AutomationBench

48.2%

#5 of 11

Reasoning & science

Humanity's Last Exam

62.5%

with tools

#6 of 42

Knowledge work

GDPval-AA v2

1769

#4 of 20

Robustness

BullshitBench v2

72%

#14 of 77

Source: BenchLM, retrieved 31 August 2026 · Source: Terminal-Bench, retrieved 30 August 2026. Every other score here is the figure the lab published at launch.

About

GLM-5.3, released August 14, 2026, was Z.ai's coding-and-cybersecurity push — "Built to Code. Ready for Cyber Defense." — built by post-training the 743B GLM-5-generation base rather than scaling it up. At release it scored 66.9% on DeepSWE, more than twenty points above GLM-5.2, with 62.5% on Humanity's Last Exam with tools and a GDPval-AA v2 rating of 1769 — at the time the best published figure from any open-model lab, and ahead of the Claude and GPT flagships in Z.ai's own launch comparison.

The cybersecurity positioning was the real novelty: 84.5% on CyberGym topped Z.ai's launch chart outright, and 54.4% on ExploitBench — CMU's exploit-development ladder — set a launch-time high among open-model labs, though the closed Anthropic and OpenAI flagships still led that test by over twenty points. Unusually for the GLM line, launch day came without weights: access started through the GLM Coding Plan and ZCode, with API access and open weights slated to follow in stages after safety evaluations. Z.ai published them on Hugging Face on August 28, 2026, two weeks after launch and two days after the smaller GLM-5.3-Flash had shipped open on day one. It was Z.ai's seventh flagship release in thirteen months.

Compare GLM-5.3 with

GLM-5.3

Suggested comparisons

GLM-5.3vsGLM-5.3-Flash GLM-5.3vsGPT-6 Astra GLM-5.3vsClaude Fable 5.1 GLM-5.3vsGemini 3.8 Flash GLM-5.3vsMuse Spark 1.3 GLM-5.3vsGrok 4.6 GLM-5.3vsDeepSeek-V4.1-Flash GLM-5.3vsMistral Medium 3.5 GLM-5.3vsKimi K3 GLM-5.3vsQwen3.8-Max-0902 GLM-5.3vsNemotron 3.5 Lightning

Frequently asked questions

: GLM-5.3 was released by Z.ai on Friday, Aug 14 2026.
: GLM-5.3 was built by Z.ai. Chinese AI lab spun out of Tsinghua University (formerly Zhipu AI), building the open-weight GLM family. Rebranded internationally as Z.ai in 2025.
: GLM-5.3 reports 13 tracked benchmark scores — BullshitBench v2: 72%; DeepSWE 1.1: 66.9%; Terminal-Bench 4.0: 41.82%; Terminal-Bench 3.0: 28.3%; Terminal-Bench 2.1: 88.2%; CyberGym: 84.5%; ExploitBench: 54.4%; ExploitGym (6-hour budget): 130; ExploitGym (2-hour budget): 105; Humanity's Last Exam (with tools): 62.5%; Agent's Last Exam (pass@1): 28.5%; AutomationBench: 48.2%; GDPval-AA v2: 1769. Scores are the figures published at release by Z.ai. It holds the best score among all models tracked here on ExploitGym (6-hour budget) and ExploitGym (2-hour budget).
: GLM-5.3 is reported at 743B parameters.
: Partly. GLM-5.3 is an open-weight model: the trained weights are free to download and run locally or on your own infrastructure, but the training data and code are not fully released and the license may restrict some commercial uses. It is not open source in the strict sense.
: Z.ai's previous tracked release was GLM-5.2 on Jun 16 2026, 59 days earlier. It was followed by GLM-5.3-Flash on Aug 26 2026.