GLM-5.3

+ contribute

GLM-5.3 is an open-weights language model from Z.ai, released August 2026. Its published size is 753B parameters (40B active per token) and a 1M-token context window. Licence: glm-5.3-license.

76.1±0.8#33 of 719 LLMsfrom 13 benchmark results across 4 sources

$ how it compares

The LLMs ranked either side of GLM-5.3, out of 828 with a score, then more from Z.ai.

#	model	score	where	RAM	released
31	DeepSeek V4 Pro (0813) DeepSeek	76±1	api $1.63/M +local	~1010GB	2026-08
32	Gemini 3.7 Flash Google	76±1	api $1.50/M		2026-08
33	GLM-5.3 Z.ai	76±1	api $2.15/M +local	~239GB	2026-08
34	Qwen3.8 Max (0902) Alibaba	76±1	api $3.00/M		2026-09
35	DeepSeek V3.2 Speciale DeepSeek	76±3	comparison only
	more from Z.ai
52	GLM-5.3-Flash Z.ai	74±1	api $0.24/M +local	~103GB	2026-08
57	GLM-5.2 Z.ai	74±1	api $2.15/M +local	~121GB	2026-06
81	GLM-5.1 Z.ai	71±1	api $2.15/M +local	~119GB	2026-04

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #14 of 300 * · 16% of its score · 2 tests

Beats GPT-5.5 Pro, GPT-5.4 Pro, GPT-5.5 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Arena Agent (October 2026 aggregate IPS)	0.0236 IPS ✎	▁▁▁▄▂▃▇█▂▂▁▃ beats 27/49	avg -0.312 IPS	▇▇▇▇ 11%	sharp
Terminal-Bench 4.0	41.8% ✎	█▅▅▅▅·▅▃▃▅▅▅ beats 13/20	very hard · avg 0.4%	▇▇ 5.7%	sharp

coding▁▁▁▃▆██▇▆▄▃▁ #21 of 458 * · 13% of its score · 2 tests

Beats GPT-5.5, MiMo-V2.6-Pro, GPT-5.6 Terra here, though they rank higher overall. Trails Grok 4.7, Qwen3.8-Flash-Next, Qwen3.8 Max (0902) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1623.0 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 102/119	avg 981	▇▇▇▇ 9.3%	sharp
Luau Leetcode fully solved (Torpedo Software)	69.7% ✎	▁▁▁▁·▁▂▃▅▃█▇ beats 23/49	hard · avg 18%	▇ 3.4%	typical

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #22 of 393 * · 5.2% of its score · 1 test

Beats GPT-6 Astra, Claude Sonnet 5.5, GPT-6 Sol here, though they rank higher overall. Trails Gemini 3.6 Flash, Claude Opus 4.6, Muse Spark 1.1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1478.5 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 356/379	avg 1268	▇▇ 5.2%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #28 of 398 · 46% of its score · 3 tests

Beats DeepSeek V4.1 Flash, Gemini 3.8 Flash, DeepSeek V4 Pro (0813) here, though they rank higher overall. Trails Qwen3.7 Max, Qwen3.8 Max (0902) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
FrontierMath tiers 1–3 v2	68.8% ✎	▂··▁▁▃▄▄█▅▃▃ beats 58/82	very hard · avg 1.2%	▇▇▇▇▇▇▇▇▇ 23%	sharp · at its level
FrontierMath tier 4 v2	29.3% ✎	▁▂▁▂█▅▂▂▁▁▁▁ beats 35/63	very hard · avg 0.1%	▇▇▇▇▇▇▇▇▇ 23%	sharp · at its level
OTIS mock AIME 2024–2025	91.1% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 148/193	very hard · avg 8.5%	▇ 0.8%	sharp · maxed out · noisy

other tests5 not in a task · 19% of its score

test	result	vs the others in its run	difficulty	counts	why
Mystery Game Puzzles (Epoch)	33.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 52/74	very hard · avg 9.4%	▇▇▇ 7.6%	sharp · noisy
Chess Puzzles (Epoch)	21.0% ✎	█▁▁▁▁▂▁▄▂▂▁▁ beats 87/135	very hard · avg 5.9%	▇▇ 4.7%	sharp · at its level · noisy
GPQA Diamond	90.9% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 181/208	hard · avg 40%	▇ 3.6%	sharp
SimpleQA Verified	41.0% ✎	▁▁▃▂▃▃▆▆█▃▂▅ beats 36/78	very hard · avg 8.0%	▇ 2.1%	at its level
RobloxQA v2.0 multiple choice (Torpedo Software)	93.9% ✎	▁▁▃▁█▄▃▃▆▂▂▂ beats 25/49	easy · avg 78%	▇ 1.2%	maxed out

$ where to run it

Through an API

api · $2.15/M · $1.4 in / $4.4 out per 1M tokens · 1M context · docs.z.ai ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

llama.cpp · recommended

version

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run GLM-5.3 · needs ~452.3 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/GLM-5.3-GGUF:UD-Q4_K_XL
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/GLM-5.3-GGUF:UD-Q4_K_XL

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run GLM-5.3 · needs ~452.3 GB · from huggingface.co/unsloth
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/unsloth/GLM-5.3-GGUF@UD-Q4_K_XL -y
Then run lms ls to see its name, and lms chat followed by that name

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run GLM-5.3 · needs ~458.9 GB, Mac only · from huggingface.co/orcarouter
chat in the terminal
mlx_lm.chat --model orcarouter/GLM-5.3-MLX
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model orcarouter/GLM-5.3-MLX

Weights and documentation: huggingface.co/zai-org ↗

$ how the score was worked out · Epoch AI — internal evaluatio… 64%, LMArena leaderboard dataset 25%, Terminal-Bench 4.0 official l… 6% and 1 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Epoch AI — internal evaluations	7	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 64%	CC BY 4.0
LMArena leaderboard dataset	3	▇▇▇▇▇▇▇▇ 25%	CC BY 4.0
Terminal-Bench 4.0 official leaderboard submissions	1	▇▇ 6%	Apache-2.0
Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode	2	▇ 5%	MIT
prior · pull toward the middle		▇ 0%
weighted average		= 76.1