GLM-4.6

+ contribute

GLM-4.6 is an open-weights language model from Z.ai, released September 2025. Its published size is 357B parameters (32B active per token) and a 205k-token context window. Licence: mit.

66.1±0.8#128 of 719 LLMsfrom 31 benchmark results across 7 sources

$ how it compares

The LLMs ranked either side of GLM-4.6, out of 828 with a score, then more from Z.ai.

#	model	score	where	RAM	released
145	Qwen3.6 Flash Alibaba	66±1	api $0.56/M		2026-04
146	MiMo-V2-Flash (309B-A15B) Xiaomi	66±1	local	~108GB	2025-12
147	GLM-4.6 Z.ai	66±1	api $1.00/M +local	~58GB	2025-09
148	Kimi K2 Instruct (0711) Moonshot AI	66±1	api $1.00/M +local	~160GB	2025-07
149	o1 (2024-12-17) OpenAI	66±1	api $26.25/M		2024-12
	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
57	GLM-5.2 Z.ai	74±1	api $2.15/M +local	~121GB	2026-06

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #109 of 300 · 41% of its score · 13 tests

Beats o3-pro, Qwen3.6 Plus, o3 here, though they rank higher overall. Trails SmolLM3 3B, Nova 2 Pro (Preview), Mistral Medium 3.5 (128B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Verified lab +1	68.0% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 68/131	hard · avg 24%	▇▇▇ 7.7%	sharp · at its level · small run · several runs
Terminal-Bench 2.0 with Terminus 2 agent (benchmark authors' runs)	24.5% ✎	▃··▃·▃▃██▅▃▅ beats 6/14	very hard · avg 5.8%	▇▇ 5.5%	sharp · at its level
BrowseComp lab +1	45.1% ✎	▂·▄▁▂▃▆▄█▂▃▄ beats 17/42	hard · avg 18%	▇▇ 4.2%	at its level · small run · several runs
Terminal-Bench Hard lab	23.6% ✎	▂▅▆▂▂▆▂▂█▅▅▂ beats 17/27	very hard · avg 4.5%	▇▇ 3.9%	sharp · small run
BFCL V4 (December 2025 archive)	72.4% ✎	▁▁▁▂██▅▅▆▂▂▃ beats 75/77	hard · avg 32%	▇ 3.1%	at its level
Terminal-Bench 2.0 lab	24.5% ✎	▂▂▅▅████▆▂▃▃ beats 4/36	very hard · avg 8.4%	▇ 3.0%	at its level · small run
τ²-bench aggregate lab	75.2% ✎	▁▁▁·▂▂▄▃▃▅█▂ beats 18/37	medium · avg 41%	▇ 2.8%	small run
BrowseComp-ZH lab	49.5% ✎	▁▁▁▁▁▃·▁▄▅▄█ beats 13/23	hard · avg 19%	▇ 2.8%	at its level · small run

5 more tests

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Multilingual lab	53.8% ✎	▁··▁▃▂▅▇█▃▃▁ beats 13/60	hard · avg 25%	▇ 2.4%	at its level · small run
Terminal-Bench (pre-2.0 release) lab	40.5% ✎	▃·▃···▃▃▅·▅█ beats 9/10	hard · avg 15%	▇ 1.8%	at its level · small run
BrowseComp with context management lab	57.5% ✎	▅·▅▃·█▃·█▅·▃ beats 0/14	medium · avg 37%	▇ 1.6%	at its level · small run
Humanity's Last Exam with tools lab +1	30.4% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 10/54	very hard · avg 15%	▇ 1.6%	at its level · small run
τ²-bench overall (weighted) lab	75.9% ✎	█··········█ beats 1/1		—	not counted yet: no other model in the same run to compare it with

coding▁▁▁▃▆██▇▆▄▃▁ #118 of 458 · 29% of its score · 6 tests

Beats Qwen3.5 Plus, Gemini 2.5 Pro, Kimi K2 Thinking here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, Solar Pro 4, Nemotron 3 Super 120B-A12B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1339.4 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 26/119	avg 981	▇▇▇▇ 11%	sharp
SWE-bench Verified lab +1	68.0% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 68/131	hard · avg 24%	▇▇▇ 7.7%	sharp · at its level · small run · several runs
LiveCodeBench v6 (window unspecified) lab +1	82.8% ✎	▂▃▄▄▅██▅▄▇▃▆ beats 48/65	medium · avg 47%	▇▇ 5.1%	small run · several runs
WebApp1K-Duo-React pass@1 (onekq-ai)	57.8% ✎	▁··▁·▁▁▁▆█▇█ beats 10/45	hard · avg 32%	▇ 2.9%	at its level
SWE-bench Multilingual lab	53.8% ✎	▁··▁▃▂▅▇█▃▃▁ beats 13/60	hard · avg 25%	▇ 2.4%	at its level · small run
LiveCodeBench v6 (window unspecified) with tools lab	84.5% ✎	no other model yet		—	not counted yet: no other model in the same run to compare it with

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #84 of 393 * · 6.3% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B), Inkling Small (266B-A12B), Qwen3.5 122B-A10B here, though they rank higher overall. Trails Gemma 4 12B, Gemma 4 26B-A4B, Mistral Medium 3.5 (128B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1424.1 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 275/379	avg 1268	▇▇▇ 6.3%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #86 of 398 · 16% of its score · 5 tests

Beats Claude Sonnet 4.6, Claude Opus 4.5, MiniMax-M3 here, though they rank higher overall. Trails Gemini 2.5 Flash, Granite 4.2 30B, INTELLECT-3 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2025 lab +1	93.9% ✎	▁·▁▁▁▂▅▆█▇▅▁ beats 128/133	medium · avg 62%	▇▇▇ 8.2%	sharp · maxed out · small run · several runs
IMO-AnswerBench lab	73.5% ✎	▁▁·▁▁▁▄▅█▄▂▃ beats 12/42	medium · avg 36%	▇ 3.4%	small run
HMMT February 2025 lab	89.2% ✎	▁▁▁▃▁▅▅▃██▁▄ beats 18/31	easy · avg 65%	▇ 2.4%	small run
HMMT November 2025 lab	87.7% ✎	▁▁····▁▅▃█▇▁ beats 9/23	easy · avg 68%	▇ 1.9%	small run
AIME 2025 with tools lab	98.6% ✎	no other model yet		—	not counted yet: no other model in the same run to compare it with

web search▂▃▂▆▄█▃▃▁▁·▁ #74 of 80 · 10% of its score · 4 tests

Beats Gemini 3.1 Flash-Lite here, though they rank higher overall. Trails GPT-4.1 nano, GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp lab +1	45.1% ✎	▂·▄▁▂▃▆▄█▂▃▄ beats 17/42	hard · avg 18%	▇▇ 4.2%	at its level · small run · several runs
BrowseComp-ZH lab	49.5% ✎	▁▁▁▁▁▃·▁▄▅▄█ beats 13/23	hard · avg 19%	▇ 2.8%	at its level · small run
BrowseComp with context management lab	57.5% ✎	▅·▅▃·█▃·█▅·▃ beats 0/14	medium · avg 37%	▇ 1.6%	at its level · small run
Humanity's Last Exam with tools lab +1	30.4% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 10/54	very hard · avg 15%	▇ 1.6%	at its level · small run

other tests7 not in a task · 18% of its score

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam lab +1	17.2% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 39/104	very hard · avg 4.4%	▇▇ 5.5%	small run · several runs
GPQA Diamond lab	81.0% ✎	▁▁▁▁▂▄▅▅███▄ beats 202/310	hard · avg 44%	▇▇ 5.4%	sharp · small run
GPQA (subset unspecified) lab	81.0% ✎	▂▁▂▂▄▃▆▃▅█▄▃ beats 95/104	hard · avg 47%	▇▇ 4.1%	sharp · at its level · small run
MMLU-Pro lab	83.2% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 154/188	medium · avg 56%	▇ 2.6%	small run
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	95.7% ✎	▁▁▂▃▄▄█▆▃▂▂▁ beats 79/140	very easy · avg 94%	▇ 0.2%	blunt · maxed out · noisy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	90.5% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 49/98	very easy · avg 90%	▇ 0.0%	blunt
GPQA (subset unspecified) with tools lab	82.9% ✎	no other model yet		—	not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · $1.00/M · $0.6 in / $2.2 out per 1M tokens · 205k 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-4.6 · needs ~214.7 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/GLM-4.6-GGUF:UD-Q4_K_XL
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/GLM-4.6-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-4.6 · needs ~214.7 GB · from huggingface.co/unsloth
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/unsloth/GLM-4.6-GGUF@UD-Q4_K_XL -y
Then run lms ls to see its name, and lms chat followed by that name

MLX (Mac)

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run GLM-4.6 · needs ~198.6 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/GLM-4.6-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/GLM-4.6-4bit

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

$ how the score was worked out · zai-org/GLM-4.7 model card 51%, zai-org/GLM-4.6 model card 20%, LMArena leaderboard dataset 17% and 4 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
zai-org/GLM-4.7 model card · lab-reported	17	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 51%	MIT
zai-org/GLM-4.6 model card · lab-reported	7	▇▇▇▇▇▇ 20%	MIT
LMArena leaderboard dataset	2	▇▇▇▇▇ 17%	CC BY 4.0
Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces — benchmark authors' own agent-model trials	1	▇▇ 6%	CC BY 4.0
Berkeley Function Calling Leaderboard result archive	1	▇ 3%	Apache-2.0
WebApp1K models leaderboard (onekq-ai)	1	▇ 3%	MIT
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	2	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 66.1

not counted (5)

zai-org/GLM-4.6 model card	AIME 2025 with tools [correct]	98.6%	fewer than three models have this test yet, or this source ran it for this model alone
zai-org/GLM-4.6 model card	GPQA (subset unspecified) with tools [correct]	82.9%	fewer than three models have this test yet, or this source ran it for this model alone
zai-org/GLM-4.6 model card	LiveCodeBench v6 (window unspecified) with tools [correct]	84.5%	fewer than three models have this test yet, or this source ran it for this model alone
zai-org/GLM-4.6 model card	τ²-bench overall (weighted) [correct]	75.9%	fewer than three models have this test yet, or this source ran it for this model alone
zai-org/GLM-4.6 model card	Humanity's Last Exam with tools [correct]	30.4%	the only model in this source's run of the test, so there is nothing to compare it with there; another run of the test counts

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json