GLM-5.2

+ contribute

GLM-5.2 is an open-weights language model from Z.ai, released June 2026. Its published size is 753B parameters (40B active per token) and a 1M-token context window. Licence: mit.

73.6±0.5#54 of 719 LLMsfrom 52 benchmark results across 12 sources

$ how it compares

The LLMs ranked either side of GLM-5.2, out of 828 with a score, then more from Z.ai.

#	model	score	where	RAM	released
55	Hunyuan Hy3 (295B-A21B) Tencent	74±1	api $0.23/M +local	~96GB	2026-07
56	Gemini 3.6 Flash Google	74±1	api $1.50/M		2026-07
57	GLM-5.2 Z.ai	74±1	api $2.15/M +local	~121GB	2026-06
58	Dola Seed 2.0 Pro ByteDance	74±2	comparison only
59	Qwen3.8 27B Alibaba	73±1	api $1.60/M +local	~18GB	2026-08
	more from Z.ai
33	GLM-5.3 Z.ai	76±1	api $2.15/M +local	~239GB	2026-08
52	GLM-5.3-Flash Z.ai	74±1	api $0.24/M +local	~103GB	2026-08
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▁▁▃▆█▆▅▅▅▄▃▁ #39 of 300 · 23% of its score · 14 tests

Beats Gemini 3.7 Flash, Qwen3.8-Flash-Next, Qwen3.8-Omni-Flash here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Arena Agent (October 2026 aggregate IPS)	0.0348 IPS ✎	▁▁▁▄▂▃▇█▂▂▁▃ beats 30/49	avg -0.312 IPS	▇▇ 4.7%	sharp
SWE-bench Verified	78.7% ✎	▁··▁·▁▂▁▄█▄▁ beats 26/29	hard · avg 19%	▇▇ 4.1%	sharp
Terminal-Bench 2.1 lab +1	81.0% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 40/56	hard · avg 23%	▇ 2.4%	small run · several runs
GameLogicBench solve rate (72 Godot game-logic tasks, Claude Code scaffold)	37.5% ✎	█▄·█·▄·██▄·▄ beats 7/11	very hard · avg 4.7%	▇ 1.8%	at its level
E2E-SWE (build a working codebase from scratch, 186 tasks) pass@1	22.2% ✎	▃··▅··▃█▃▅▃▅ beats 1/12	very hard · avg 3.1%	▇ 1.6%	sharp · at its level
τ³-bench banking_knowledge (v1.0.1 grading)	37.1% ✎	▃▃▂▃·▆▃▂█▅▅▂ beats 15/25	very hard · avg 4.1%	▇ 1.6%	at its level
MCP-Atlas (public set) lab	76.8% ✎	█··█·██·████ beats 6/7	hard · avg 29%	▇ 1.4%	at its level
Humanity's Last Exam with tools lab +1	54.7% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 38/54	hard · avg 16%	▇ 1.3%	at its level · small run · several runs

6 more tests

test	result	vs the others in its run	difficulty	counts	why
Legal Research Bench (agentic, all-pass)	31.2% ✎	▄█··▄▄▄██·▄█ beats 9/12	very hard · avg 5.8%	▇ 0.8%	at its level
EEGAgentBench overall (tool-based EEG analysis agents, 6 tasks)	45.5% ✎	▂·▁▁··▇▇█▂·▁ beats 10/25	hard · avg 29%	▇ 0.7%	blunt · at its level
Toolathlon (Tool Decathlon) lab	48.2% ✎	▂▂▄▂▄▄▆▂█▆██ beats 13/27	very hard · avg 13%	▇ 0.7%	at its level · small run
Agents' Last Exam (metric/harness unspecified) lab	20.4% ✎	▂·▃·█▂···▂▄▄ beats 1/11	very hard · avg 6.8%	▇ 0.5%	at its level · small run
DeepSWE v1.1 (resolved) lab	46.2% ✎	▂····▁▃▁▆██▆ beats 3/28	hard · avg 19%	▇ 0.5%	at its level · small run
SWE-bench Pro lab	62.1% ✎	▅▃▃·▃·▅▅▅█▃▃ beats 15/15	hard · avg 31%	▇ 0.3%	blunt · at its level · small run

coding▁▁▁▃▆██▇▆▄▃▁ #26 of 458 · 14% of its score · 7 tests

Beats GPT-5.5, GPT-5.6 Terra, Claude Opus 4.8 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1605.0 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 96/119	avg 981	▇▇ 4.1%	sharp
SWE-bench Verified	78.7% ✎	▁··▁·▁▂▁▄█▄▁ beats 26/29	hard · avg 19%	▇▇ 4.1%	sharp
GameLogicBench solve rate (72 Godot game-logic tasks, Claude Code scaffold)	37.5% ✎	█▄·█·▄·██▄·▄ beats 7/11	very hard · avg 4.7%	▇ 1.8%	at its level
Luau Leetcode fully solved (Torpedo Software)	74.0% ✎	▁▁▁▁·▁▂▃▅▃█▇ beats 25/49	hard · avg 18%	▇ 1.7%	at its level
E2E-SWE (build a working codebase from scratch, 186 tasks) pass@1	22.2% ✎	▃··▅··▃█▃▅▃▅ beats 1/12	very hard · avg 3.1%	▇ 1.6%	sharp · at its level
DeepSWE v1.1 (resolved) lab	46.2% ✎	▂····▁▃▁▆██▆ beats 3/28	hard · avg 19%	▇ 0.5%	at its level · small run
SWE-bench Pro lab	62.1% ✎	▅▃▃·▃·▅▅▅█▃▃ beats 15/15	hard · avg 31%	▇ 0.3%	blunt · at its level · small run

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #25 of 393 * · 2.3% of its score · 1 test

Beats Claude Sonnet 5.5, GPT-6 Sol, GPT-5.6 Terra here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1475.8 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 353/379	avg 1268	▇ 2.3%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #42 of 398 · 51% of its score · 18 tests

Beats Grok 4.7, GLM-5.3-Flash, Gemini 3.6 Flash here, though they rank higher overall. Trails Kimi K2.6 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
FrontierMath tiers 1–3 v2	59.2% ✎	▂··▁▁▃▄▄█▅▃▃ beats 46/82	very hard · avg 1.2%	▇▇▇▇▇ 11%	sharp · at its level
FrontierMath tier 4 v2	29.3% ✎	▁▂▁▂█▅▂▂▁▁▁▁ beats 33/63	very hard · avg 0.1%	▇▇▇ 7.9%	sharp
MathArena ArXivMath June 2026 (48 problems)	46.5% ✎	▂···▄·██▆▆▄▂ beats 3/19	very hard · avg 4.6%	▇▇▇ 6.4%	sharp · at its level
MathArena ArXivMath April 2026 (40 problems)	47.5% ✎	▁·····▅██▄·▁ beats 6/20	very hard · avg 7.2%	▇▇ 5.3%	sharp · at its level
MathArena ArXivMath February 2026 (32 problems)	46.1% ✎	▁▁▁▁█▇▄▃▁▃·▃ beats 17/24	very hard · avg 9.1%	▇ 3.0%	at its level
MathArena ArXivMath March 2026 (30 problems)	61.7% ✎	█▃▅·█▃▃·▅·▃▃ beats 9/14	very hard · avg 15%	▇ 2.6%	at its level
MathArena ArXivMath May 2026 (40 problems)	56.7% ✎	▁··▁▄█▄▁·▃·▁ beats 9/17	very hard · avg 10%	▇ 1.9%	at its level
MathArena BrokenArXiv April 2026 (61 problems)	18.4% ✎	▂▂▆█▃▃▂▂▂··▂ beats 4/18	very hard · avg 0.7%	▇ 1.9%	sharp · at its level · noisy

10 more tests

test	result	vs the others in its run	difficulty	counts	why
MathArena BrokenArXiv June 2026 (54 problems)	14.2% ✎	▄··█▂▂▂▆▄··▂ beats 4/14	very hard · avg 1.1%	▇ 1.4%	sharp · at its level · noisy
HMMT February 2026 lab	92.5% ✎	▃▃▂▃·██▆█▅▃▅ beats 30/33	hard · avg 32%	▇ 1.3%	sharp · small run
MathArena ArXivLean June 2026 (46 problems)	8.7% ✎	▄····▄▄███·▄ beats 1/9	very hard · avg 0.7%	▇ 1.2%	sharp
MathArena BrokenArXiv May 2026 (50 problems)	10.5% ✎	██·▄▄▄····▄▄ beats 2/8	very hard · avg 1.6%	▇ 1.2%	sharp
MathArena BrokenArXiv March 2026 (56 problems)	12.1% ✎	█·▆█▂·▂▂···▂ beats 5/14	very hard · avg 0.8%	▇ 1.1%	sharp · noisy
IMO-AnswerBench lab	91.0% ✎	▁▁·▁▁▁▄▅█▄▂▃ beats 39/42	medium · avg 39%	▇ 1.1%	small run
MathArena BrokenArXiv February 2026 (31 problems)	10.5% ✎	▃▂▂▃█▃▂·▃··▂ beats 5/16	very hard · avg 0.7%	▇ 1.1%	sharp · noisy
AIME 2026 lab	99.2% ✎	▁▁▁▂▁▂▆█▆▂·▃ beats 49/49	easy · avg 66%	▇ 1.0%	sharp · maxed out · small run
HMMT November 2025 lab	94.4% ✎	▁▁····▁▅▃█▇▁ beats 21/23	easy · avg 70%	▇ 0.7%	maxed out · small run
OTIS mock AIME 2024–2025	86.4% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 128/193	very hard · avg 8.5%	▇ 0.6%	sharp · maxed out · noisy

web search▂▁▂▆▄█▃▃▁▁·▁ #39 of 80 · 2.1% of its score · 2 tests

Beats Gemini 3.1 Pro (Preview), Qwen3.7 Max, Gemini 3.5 Flash here, though they rank higher overall. Trails GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503), Llama 4 Scout (109B-A17B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam with tools lab +1	54.7% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 38/54	hard · avg 16%	▇ 1.3%	at its level · small run · several runs
Legal Research Bench (agentic, all-pass)	31.2% ✎	▄█··▄▄▄██·▄█ beats 9/12	very hard · avg 5.8%	▇ 0.8%	at its level

other tests14 not in a task · 18% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond +1	91.9% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 187/208	hard · avg 40%	▇▇ 3.9%	sharp · several runs
EBR-bench (Epoch)	9.5% ✎	▄▂·▆█▂▆▄▆▆·▄ beats 5/23	very hard · avg 0.3%	▇ 3.7%	sharp
Mystery Game Puzzles (Epoch)	15.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 22/74	very hard · avg 9.4%	▇ 2.5%	sharp · noisy
Chess Puzzles (Epoch)	21.0% ✎	█▁▁▁▁▂▁▄▂▂▁▁ beats 86/135	very hard · avg 5.9%	▇ 1.8%	sharp · noisy
Humanity's Last Exam lab	40.5% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 90/104	very hard · avg 5.8%	▇ 1.0%	at its level · small run
Legal Research Bench (single-shot, no tools, all-pass)	3.4% ✎	▃▃█▃·▅···█▃▃ beats 5/12	very hard · avg 0.8%	▇ 1.0%	too hard for it
SimpleQA Verified	34.2% ✎	▁▁▃▂▃▃▆▆█▃▂▅ beats 28/78	very hard · avg 8.0%	▇ 0.9%	at its level
NL2Repo lab	48.9% ✎	▂▂·▂▂▄█▆▆▃▄▆ beats 16/21	very hard · avg 11%	▇ 0.9%	at its level · small run

6 more tests

test	result	vs the others in its run	difficulty	counts	why
AutomationBench v1.0.6 lab	26.2% ✎	▃····▃·▃▃··█ beats 2/6	very hard · avg 5.7%	▇ 0.7%	at its level · small run
RobloxQA v2.0 multiple choice (Torpedo Software)	89.0% ✎	▁▁▃▁█▄▃▃▆▂▂▂ beats 9/49	easy · avg 78%	▇ 0.6%	typical
CritPt lab	20.9% ✎	▄···▄▄▄▄█▄▄▄ beats 9/9	very hard · avg 1.4%	▇ 0.5%	noisy · small run
WMT25 Terminology Translation (term success rate)	78.2% ✎	▂·▂··▃·▄▂·█▆ beats 2/12	medium · avg 64%	▇ 0.3%	noisy
WMT24++	78.8% ✎	█·▃·▅▅▃···▃▃ beats 7/10	easy · avg 71%	▇ 0.2%	blunt
WMT26 General MT (GEMBA-ESA, LLM judge)	76.5% ✎	▄·▄▄▄·▄██▄█▄ beats 5/12	easy · avg 65%	▇ 0.1%	blunt

$ 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-5.2 · needs ~452.3 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/GLM-5.2-GGUF:UD-Q4_K_XL
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/GLM-5.2-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-5.2 · 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-5.2-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-5.2 · needs ~418.3 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/GLM-5.2-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/GLM-5.2-4bit

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

$ how the score was worked out · Epoch AI — internal evaluatio… 35%, MathArena ArXivMath / BrokenA… 27%, zai-org/GLM-5.2 model card 12% and 9 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	9	▇▇▇▇▇▇▇▇▇▇▇ 35%	CC BY 4.0
MathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets	11	▇▇▇▇▇▇▇▇ 27%	CC BY-SA 4.0
zai-org/GLM-5.2 model card · lab-reported	13	▇▇▇▇ 12%	mit
LMArena leaderboard dataset	3	▇▇▇ 11%	CC BY 4.0
zai-org/GLM-5.3-Flash model card · lab-reported	5	▇ 4%	MIT
Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode	2	▇ 2%	MIT
Legal Research Bench (Vals AI, arXiv:2610.00609v1)	2	▇ 2%	CC-BY-4.0
GameLogicBench: Evaluating Coding Agents on Runtime Game Logic with Tick-Level State Assertions (arXiv:2609.21562v2)	1	▇ 2%	CC BY 4.0
E2E-SWE: Benchmarking LLMs on Building Working Codebases from Scratch (arXiv:2609.38335v1)	1	▇ 2%	CC BY 4.0
τ²/τ³-bench leaderboard: Sierra's own evaluation runs	1	▇ 2%	MIT
EEGAgentBench: Benchmarking LLM Agents on Short- and Long-Horizon EEG Analysis (arXiv:2609.31632v1)	1	▇ 1%	CC BY 4.0
Cohere — North Small Translate (arXiv:2609.13916v1)	3	▇ 1%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 73.6