GLM-4.6

zai-org MIT out 13 months ago

357B parameters (8 of 160 experts per token), text in, text out, 198K tokens of context. 5 providers serve it from $0.40 per million input tokens.

Verified 2026-10-25 Rank #86 in LLMs · score 101.9 method v1 Sources 4 Downloads 30d 22.4k · #447 of 619 tracked

Changed this week: 1 new provider: Z.ai.2026-10-24

Can I run it?214GB at INT4Estimateneeds a multi-GPU nodemethod

What does it cost?$0.40/1M incheapest of 4DeepInfrafp42026-10-25

Can I use it?MITPermissivelicence file2026-10-24

Does it work?72.38on BFCL#1 of 28 measuredleaderboard2026-10-25

Do people use it?22.4kdownloads, 30 days#447 of 619 we track · below the median+6 likes, downloads +13% this week1,245 likesHugging Face2026-10-24

Worth reading

Picked by rule from 16 tracked pieces: the lab's own post, the most shared, a field report, the latest.

GLM-4.6V accepts images and documents directly as tool parameters without converting them to text first.

From GLM-4.6V · z.ai · 2025-12-08

The lab's own post

GLM-4.6: Advanced Agentic, Reasoning and Coding Capabilies

Z.ai released GLM-4.6, an updated flagship model with a 200K token context window and improved coding performance.

z.ai 2025-09-30 43 HN pts · 10 comments 484 words

Most discussed · 194 HN pts

Cerebras Code now supports GLM 4.6 at 1000 tokens/sec

cerebras.ai 2025-11-08 194 HN pts · 129 comments 110 words

Latest · 11 months ago

GLM-4.6V

Z.ai officially introduces the GLM-4.6V series, which integrates native Function Calling capabilities for the first time.

GLM-4.6V accepts images and documents directly as tool parameters without converting them to text first.
The model processes approximately 150 pages of complex documents or a one-hour-long video in a single inference pass.

z.ai 2025-12-08 8 HN pts 1.4k words

16 best of 16 tracked

The 5 picks above, plus these 11. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleGLM-4.6 - Overview - Z.AI DEVELOPER DOCUMENTdocs.z.ai · undated · official
ArticleGLM-4.6: Deep Dive Into Zhipu AI's MoE Architecture - ExpertBeaconexpertbeacon.com · undated
ArticleGLM-4.6: Specs, Pricing and Open Weightsglm-ai.chat · undated
ArticleGLM-4.6 vs Claude Sonnet: A Performance & Cost Analysiscirra.ai · undated
ArticleGLM-4.6: An Open-Source AI for Coding vs. Sonnet & GPT-5intuitionlabs.ai · undated
ArticleClaude Opus 4.5 vs. GLM-4.6llm-stats.com · 2026-01-07
ArticleUsing GLM-4.6 to reverse engineer Catan Universe browser game (WebGL and unity)ankitmaloo.com · 2025-10-30
ArticleClaude Haiku 4.5 vs. GLM-4.6 vs. GPT-5 Mini: Job Queue System Benchmarkblog.kilocode.ai · 2025-10-21
ArticleGLM 4.6 code generation: a deep dive into accuracy and speed for DevOps - EONSReonsr.com · 2026-01-05
ArticleGLM-4.6: Feature, architecture and Benchmark performance - CometAPIcometapi.com · 2025-10-06
ArticleZ.AI: GLM 4.6 by Z-Ai - AI Model Details | LLMIndex | LLMIndexllmindex.net · 2025-09-29

Show all 11

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

4 providers from $0.40

4 providers with a live price, from $0.40 to $0.60 per million input tokens; the median is $0.49. One dot per provider, cheapest in teal.

#	Cheapest 5	In /1M	Out /1M	Context	Quant	Seen
1	DeepInfra	$0.40	$1.60	198K	fp4	Oct 25
2	Venice	$0.43	$1.75	193K	fp4	Oct 24
3	Novita	$0.55	$2.20	200K	bf16	Oct 25
4	Z.ai	$0.60	$2.20	198K	fp4	Oct 25
5	Featherless	—	—	—	—	Oct 25

All 5 providers →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Widely served (5 API providers); 38 quantized builds for local use.

As a service5DeepInfra, Featherless, Novita, Venice +1
On own hardware38quantized builds
Built on2522 fine-tunes · 3 adapters · 0 merges
In apps100+Hugging Face Spaces

Top Spaces running it: akhaliq/anycoder ♥ 3,321 · hadadxyz/ai ♥ 819 · umint/ai ♥ 258 · dropkickJesus999/deepsiteV3 ♥ 32

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Needs an 8×80 GB node even at INT4

About 214 GB at INT4, 428 GB at FP8, 856 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.

Estimate params × bytes per param × 1.2 · how we estimate

How it scores

Third-party measurements only; author-reported numbers stay in Reference, marked as such.

#1 of 28 on BFCL so far

72.38. Each tick is a model on this site with a measured BFCL row; author-reported numbers stay in Reference. Percent (overall accuracy).

Source · leaderboard · 2026-10-25

The story so far

Weights, every dated article, and repo commits on one rail.

3 dated articles and 12 repo commits so far

Teal is the weights (2025-09-29), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.

What people built on it

63 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

63 derivatives on the Hub, most of them quantized

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF unsloth/GLM-4.6-GGUF for llama.cpp, Ollama, LM Studio 256.5k downloads MLX mlx-community/GLM-4.6-4bit for Apple silicon 21.9k downloads AWQ QuantTrio/GLM-4.6-AWQ for vLLM, SGLang 93.5k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

3 rows

Benchmarks 3 rows

measured third-party leaderboard row (3)author-reported number from the model card or eval filesNo confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
measured BFCL · Call tools and APIs	72.38 percent (overall accuracy)	5 sub-scores Non-Live AST Acc: 87.56 Live Acc: 80.9 Multi Turn Acc: 68 Web Search Acc: 77.5 Memory Acc: 55.7	leaderboard	2026-10-25
measured SWE-bench Verified · Write code with agents	68.2 percent resolved	agent: Undisclosed	leaderboard	2025-09-30
measured SWE-bench Verified · Write code with agents	55.4 percent resolved	agent: mini-SWE-agent	leaderboard	2025-12-01

Use-case evidence

5 use cases

What it is good at 5 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

24 fields

Listed in LLMs · MoE

Facts

GLM-4.6 has 357B parameters. (safetensors metadata, observed 2026-10-24)
GLM-4.6 is a mixture-of-experts model with 160 experts, routing 8 per token. (config.json, observed 2026-10-24)
Context length is 202,752 tokens (198K). (config.json max_position_embeddings, observed 2026-10-24)
License: MIT (friction: none). Permissive. (license file, observed 2026-10-24)
Created 2025-09-29, last modified 2025-09-30. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	Glm4MoeForCausalLM
Model type	glm4_moe
Dtype	bfloat16
Params by dtype	BF16: 356,785,884,416 · F32: 14,400
Size band	XXL
MoE	yes
Experts	160
Experts per token	8
Context length	202,752 (config.json max_position_embeddings)
Languages	en, zh
Papers	arXiv:2508.06471
License	MIT (mit)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Glm4MoeForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Glm4MoeForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for GLM-4.6: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	856 GB · Cluster (≥640 GB)
FP8 (1 byte/param)	428 GB · Multi-GPU node (<640 GB)
INT4 (0.5 bytes/param)	214 GB · Multi-GPU node (<640 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.

All variants

40 variants

Lineage

Original model Family: zai-org/GLM-4.6 63 derivatives on the Hub

quantized: 38finetune: 22adapter: 3

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
unsloth/GLM-4.6-GGUF	quantized	GGUF	—	—	mit	unsloth	152	256.5k
QuantTrio/GLM-4.6-AWQ	quantized	AWQ	50.4B	XL	mit	QuantTrio	5	93.5k
bartowski/zai-org_GLM-4.6-GGUF	quantized	GGUF	—	—	—	bartowski	19	44.4k
ArliAI/GLM-4.6-Derestricted-v3	finetune	—	357B	XXL	mit	ArliAI	155	16.8k
mlx-community/GLM-4.6-4bit	quantized	-4bit	353B	XXL	mit	mlx-community	15	21.9k
ReadyArt/GLM-4.6-Derestricted-GGUF	quantized	GGUF	—	—	—	ReadyArt	28	21.3k
ubergarm/GLM-4.6-GGUF	quantized	GGUF	—	—	mit	ubergarm	28	13.9k
bullpoint/GLM-4.6-AWQ	quantized	AWQ	51.5B	XL	mit	bullpoint	5	13.2k
Downtown-Case/GLM-4.6-128GB-RAM-IK-GGUF	quantized	GGUF	—	—	mit	Downtown-Case	19	11.1k
gghfez/GLM-4.6-control-vectors	adapter	GGUF	—	—	apache-2.0	gghfez	7	7.9k
DevQuasar/zai-org.GLM-4.6-GGUF	quantized	GGUF	—	—	—	DevQuasar	1	6.3k
lukealonso/GLM-4.6-NVFP4	quantized	NVFP4	177B	XXL	mit	lukealonso	6	6k
mlx-community/GLM-4.6-5bit	quantized	—	353B	XXL	mit	mlx-community	3	4k
RedHatAI/GLM-4.6-NVFP4	quantized	NVFP4	199B	XXL	—	RedHatAI	0	3.8k
lovedheart/GLM-4.6-GGUF-IQ1_M	quantized	GGUF	—	—	mit	lovedheart	2	3.6k
unsloth/GLM-4.6	finetune	—	357B	XXL	mit	unsloth	7	2.2k
QuantTrio/GLM-4.6-GPTQ-Int4-Int8Mix	quantized	GPTQ	69.4B	XL	mit	QuantTrio	4	2.2k
mlx-community/GLM-4.6-bf16	finetune	-bf16	353B	XXL	mit	mlx-community	3	2k
RESMP-DEV/GLM-4.6-NVFP4	quantized	NVFP4	199B	XXL	mit	RESMP-DEV	5	1.8k
mgoin/GLM-4.6-FP8-BLOCK	quantized	-FP8	357B	XXL	mit	mgoin	0	2k
sm54/GLM-4.6-MXFP4_MOE	quantized	GGUF	—	—	—	sm54	8	1.4k
Wwayu/GLM-4.6-mlx-2Bit	quantized	mlx	353B	XXL	mit	Wwayu	0	1.6k
cerebras/GLM-4.6-REAP-268B-A32B	finetune	—	269B	XXL	mit	cerebras	12	844
jesusoctavioas/GLM-4.6-mlx-4Bit	quantized	mlx	353B	XXL	mit	jesusoctavioas	0	1.4k
INC4AI/GLM-4.6-gguf-q2ks-mixed-AutoRound	quantized	gguf	—	—	—	INC4AI	6	830
finding1/GLM-4.6-MLX-8.5bpw	quantized	MLX	353B	XXL	mit	finding1	0	1.1k
cs2764/GLM-4.6-mlx-6Bit	quantized	mlx	353B	XXL	mit	cs2764	0	993
andyjack/GLM-4.6-REAP-252B-A32B-MXFP4_MOE_GGUF	quantized	GGUF	—	—	mit	andyjack	0	890
saviochow/GLM-4.6-mlx-2Bit	quantized	mlx	353B	XXL	mit	saviochow	1	770
RedHatAI/GLM-4.6-quantized.w4a16	quantized	-quant	48.3B	XL	—	RedHatAI	0	777
MikeRoz/GLM-4.6-exl3	quantized	exl3	—	—	mit	MikeRoz	11	220
anikifoss/GLM-4.6-HQ4_K	quantized	GGUF	—	—	mit	anikifoss	2	523
cerebras/GLM-4.6-REAP-252B-A32B	finetune	—	252B	XXL	mit	cerebras	4	397
RedHatAI/GLM-4.6-FP8-dynamic	quantized	-FP8	353B	XXL	—	RedHatAI	0	566
moxin-org/GLM-4.6-GGUF	quantized	GGUF	—	—	mit	moxin-org	3	379
tiobrown59/GLM-4.6-Derestricted-v3	finetune	—	357B	XXL	mit	tiobrown59	0	517
gghfez/GLM-4.6-REAP-266B-A32B-Q4_K	finetune	Q4_K	—	—	—	gghfez	3	330
gghfez/GLM-4.6-REAP-266B-A32B-Q2_K	finetune	Q2_K	—	—	—	gghfez	2	371
mratsim/GLM-4.6-EXL3	quantized	EXL3	—	—	mit	mratsim	4	269
RedHatAI/GLM-4.6-quantized.w8a8	quantized	-quant	353B	XXL	—	RedHatAI	0	424

Repository changelog

12 commits

changed in the last 7 daysNew providers: 2026-10-24Rank band moves: 2026-10-24, 2026-10-23all changes

Repository changelog 12 commits

Date	Commit	SHA
2025-09-30	update `README.md`	be72194
2025-09-30	Update README.md	7cf41c1
2025-09-30	Update generation_config.json	2e51388
2025-09-30	fix: updates `README.md`	821e3e5
2025-09-30	Update README.md	ba9b1e0
2025-09-30	Update README.md	b5b1c98
2025-09-30	Update README.md	d9a318f
2025-09-30	Update README.md	e0efc5b
2025-09-30	Update config.json	966b647
2025-09-29	Add files using upload-large-folder tool	3d93363
2025-09-29	Add files using upload-large-folder tool	770364b
2025-09-29	initial commit	22ccb16

All providers

5 providers · 5 endpoints

Where to run 5 providers · 5 endpoints

Hosted endpoints observed for GLM-4.6, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.40/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
DeepInfra hf, deepinfra, openrouter	zai-org/GLM-4.6	198K	fp4	$0.40	$1.60	33.8	334 ms	yes	live	2026-10-25
Venice openrouter	z-ai/glm-4.6	193K	fp4	$0.43	$1.75	—	—	yes	live	2026-10-24
Novita hf, openrouter	zai-org/glm-4.6	200K	bf16	$0.55	$2.20	26.2	4189 ms	yes	live	2026-10-25
Z.ai hf, openrouter	glm-4.6	198K	fp4	$0.60	$2.20	38.3	2596 ms	yes	live	2026-10-25
Featherless hf	zai-org/GLM-4.6	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

16 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official3

GLM-4.6 - Overview - Z.AI DEVELOPER DOCUMENT

docs.z.aiofficialauthority570 words

Z.AI released GLM-4.6 with a 200K token context window to handle complex agentic tasks and real-world coding. The model surpassed Claude Sonnet 4 in 74 real-world coding tests within the Claude Code environment. It is over 30% more efficient than GLM-4.5, achieving the lowest consumption rate among comparable models.written, reviewed, cited

GLM-4.6: Advanced Agentic, Reasoning and Coding Capabilies

z.ai2025-09-30officialauthorityHN 43 pts · 10 comments484 words

Z.ai released GLM-4.6, an updated flagship model with a 200K token context window and improved coding performance. The model holds competitive advantages over DeepSeek-V3.2-Exp and Claude Sonnet 4 but lags behind Claude Sonnet 4.5 in coding ability. It achieves near parity with Claude Sonnet 4 in real-world tasks while using 15% fewer tokens than GLM-4.5. The GLM Coding Plan offers 1/7th the price with 3x the usage quota, and local deployment supports vLLM and SGLang.written, reviewed, cited

GLM-4.6V

z.ai2025-12-08officialauthorityHN 8 pts · 0 comments1,383 words

Z.ai officially introduces the GLM-4.6V series, which integrates native Function Calling capabilities for the first time. The release includes GLM-4.6V (106B) for cloud scenarios and GLM-4.6V-Flash (9B) for local deployment, both scaling context to 128k tokens. The model achieves SOTA performance on benchmarks including MMBench, MathVista, and OCRBench among open-source models of comparable scale.written, reviewed, cited

GLM-4.6V accepts images and documents directly as tool parameters without converting them to text first.
The model processes approximately 150 pages of complex documents or a one-hour-long video in a single inference pass.
The team extends the Model Context Protocol to use URLs for identifying multimodal content passed to and from tools.

Analysis7

Cerebras Code now supports GLM 4.6 at 1000 tokens/sec

cerebras.ai2025-11-08HN front pageHN 194 pts · 129 comments110 words

GLM-4.6 (Reasoning) Intelligence, Performance & Price Analysis

artificialanalysis.aiauthority939 words

GLM-4.6 (Reasoning) scores 19 on the Artificial Analysis Intelligence Index, placing it below average among comparable models. The model generates output at 33.4 tokens per second, which is at the lower end compared to other open weight models of similar size.generated, cited

GLM 4.6 Explained: Z.AI's Prior-Gen Coding Model

layer3labs.io2,320 words

GLM-4.6: Deep Dive Into Zhipu AI's MoE Architecture - ExpertBeacon

expertbeacon.com

GLM-4.6: Specs, Pricing and Open Weights

glm-ai.chat5,410 words

GLM-4.6 vs Claude Sonnet: A Performance & Cost Analysis

cirra.ai7,880 words

GLM-4.6: An Open-Source AI for Coding vs. Sonnet & GPT-5

intuitionlabs.ai6,437 words

All coverage tracked for this model (16) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
GLM-4.6 - Overview - Z.AI DEVELOPER DOCUMENT	docs.z.ai official	—	2026-10-07	exa
GLM-4.6 (Reasoning) Intelligence, Performance & Price Analysis	artificialanalysis.ai	—	2026-10-07	exa
GLM 4.6 Explained: Z.AI's Prior-Gen Coding Model	layer3labs.io	—	2026-10-07	exa
GLM-4.6: Deep Dive Into Zhipu AI's MoE Architecture - ExpertBeacon	expertbeacon.com	—	2026-10-07	exa
GLM-4.6: Specs, Pricing and Open Weights	glm-ai.chat	—	2026-10-07	exa
GLM-4.6 vs Claude Sonnet: A Performance & Cost Analysis	cirra.ai	—	2026-10-07	exa
GLM-4.6: An Open-Source AI for Coding vs. Sonnet & GPT-5	intuitionlabs.ai	—	2026-10-07	exa
Claude Opus 4.5 vs. GLM-4.6	llm-stats.com	2026-01-07	2026-10-07	hn
GLM 4.6 code generation: a deep dive into accuracy and speed for DevOps - EONSR	eonsr.com	2026-01-05	2026-10-07	exa
GLM-4.6V	z.ai official	2025-12-08	2026-10-07	hn
Cerebras Code now supports GLM 4.6 at 1000 tokens/sec	cerebras.ai	2025-11-08	2026-10-07	hn
Using GLM-4.6 to reverse engineer Catan Universe browser game (WebGL and unity)	ankitmaloo.com	2025-10-30	2026-10-07	hn
Claude Haiku 4.5 vs. GLM-4.6 vs. GPT-5 Mini: Job Queue System Benchmark	blog.kilocode.ai	2025-10-21	2026-10-07	hn
GLM-4.6: Feature, architecture and Benchmark performance - CometAPI	cometapi.com	2025-10-06	2026-10-07	exa
GLM-4.6: Advanced Agentic, Reasoning and Coding Capabilies	z.ai official	2025-09-30	2026-10-07	hn
Z.AI: GLM 4.6 by Z-Ai - AI Model Details \| LLMIndex \| LLMIndex	llmindex.net	2025-09-29	2026-10-07	exa

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models this model

GLM-4. 5 Team, Aohan Zeng, Xin Lv +168 · 2025-08-08 · PDF · HF paper page (217 upvotes)

Similar models

3 models

Similar models

GLM-4.7 · 358B
same size band (XXL), also MoE, same task (text-generation), fewer providers, permissive license
DeepSeek-V4-Flash-0731 · 304B
same size band (XXL), also MoE, same task (text-generation), more providers, permissive license
GLM-4.5 · 358B
same size band (XXL), also MoE, same task (text-generation), fewer providers, permissive license

Sources

4 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:34 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:01 UTC · Last verified 2026-10-25

Follow GLM-4.6

Watch this model

/feeds/models/zai-org--GLM-4.6.xml(RSS 2.0) · every change to GLM-4.6
/feeds/families/zai-org--GLM-4.6.xml(RSS 2.0) · every change across the GLM-4.6 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.