# GLM-4

> Source: https://aiwiki.ai/wiki/glm_4
> Updated: 2026-06-24
> Categories: AI Models, Chinese AI, Large Language Models
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "GLM-4." aiwiki.ai, 24 Jun 2026. https://aiwiki.ai/wiki/glm_4
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

GLM-4 is the fourth-generation [foundation model](https://aiwiki.ai/wiki/foundation_models) family from [Zhipu AI](https://aiwiki.ai/wiki/zhipu_ai), a Beijing company spun out of the Knowledge Engineering Group at [Tsinghua University](https://aiwiki.ai/wiki/tsinghua_university) and now trading internationally as [Z.ai](https://aiwiki.ai/wiki/z_ai). First announced on 16 January 2024, the flagship GLM-4 was pre-trained on roughly ten trillion tokens and, by its makers' measurements, "closely rivals or outperforms GPT-4" on general benchmarks such as [MMLU](https://aiwiki.ai/wiki/mmlu), [GSM8K](https://aiwiki.ai/wiki/gsm8k) and [HumanEval](https://aiwiki.ai/wiki/humaneval) while outperforming [GPT-4](https://aiwiki.ai/wiki/gpt-4) on Chinese alignment [1][2][3]. The family spans closed API models served through the bigmodel.cn platform, an agentic "All Tools" mode, the multimodal GLM-4V, the 2024 flagship [GLM-4-Plus](https://aiwiki.ai/wiki/glm_4_plus), and an open-weights GLM-4-9B series whose lineage of open models had drawn more than 10 million downloads on [Hugging Face](https://aiwiki.ai/wiki/hugging_face) in 2023 alone [3]. GLM-4 was first announced at the company's inaugural Technology Open Day (Zhipu DevDay) and is documented in the technical report "ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 (All Tools)" ([arXiv](https://aiwiki.ai/wiki/arxiv):2406.12793) [1][2][3].

GLM-4 continues the lineage that runs from the [GLM-130B](https://aiwiki.ai/wiki/glm_130b) base model through the [ChatGLM](https://aiwiki.ai/wiki/chatglm) chat series, all built on the General [Language Model](https://aiwiki.ai/wiki/language_model) (GLM) autoregressive blank-infilling architecture co-developed by [Jie Tang](https://aiwiki.ai/wiki/tang_jie) and colleagues. According to the technical report, the GLM-4 models were pre-trained on roughly ten trillion tokens, mostly Chinese and English, plus a smaller corpus drawn from 24 languages, then aligned through [supervised fine-tuning](https://aiwiki.ai/wiki/supervised_fine-tuning) and [reinforcement learning](https://aiwiki.ai/wiki/reinforcement_learning) from human feedback [3]. It precedes the [GLM-4.5](https://aiwiki.ai/wiki/glm_4_5) and [GLM-4.6](https://aiwiki.ai/wiki/glm_4_6) models, which are covered separately.

## What is the GLM architecture and how does it relate to ChatGLM?

The "GLM" in GLM-4 stands for General Language Model, the autoregressive blank-infilling pretraining objective Zhipu researchers introduced in 2021 and later scaled to the 130-billion-parameter GLM-130B base model. The consumer-facing chat line began with ChatGLM-6B, a 6.2-billion-parameter bilingual chat model open-sourced on 14 March 2023, trained on about 1 trillion tokens of Chinese and English and small enough to run locally on a consumer GPU using INT4 quantization [3]. GLM-4 is the direct successor to that lineage: the technical report states it represents "our most capable models that are trained with all the insights and lessons gained from the preceding three generations of ChatGLM" [3]. The report also notes that Zhipu's open models, including ChatGLM-6B (three generations), GLM-4-9B, GLM-4V-9B, WebGLM and [CodeGeeX](https://aiwiki.ai/wiki/codegeex), attracted "over 10 million downloads on Hugging face in the year 2023 alone" [3].

## API model tiers

Zhipu serves several GLM-4 variants through its API. The first release, internally dated 0116, was followed by an improved GLM-4 (0520) and a set of smaller, cheaper tiers introduced through 2024. Zhipu has described the lineup as covering price points roughly 10x, 100x and 1000x below the flagship while keeping usable quality [4]. The main tiers are summarized below.

| Tier | Context length | Notes |
| --- | --- | --- |
| GLM-4 | 128K | Flagship; versions dated 0116 and 0520 [3][5] |
| GLM-4-Air | 128K | Cost-performance tier (version 0605), performance close to GLM-4 (0116) with lower latency and cost [4][5] |
| GLM-4-AirX | 8K | High-performance Air variant, about 2.6x faster inference than GLM-4-Air [4] |
| GLM-4-Flash | 128K | Lightweight tier, made free to the public on 27 August 2024; stable throughput around 72 [tokens per second](https://aiwiki.ai/wiki/tokens_per_second) [6] |
| GLM-4-Long | 1M | Ultra-long-context tier launched 13 August 2024, priced as low as 0.001 yuan per 1,000 tokens [7] |
| GLM-4-Plus | 128K | 2024 flagship update, announced 29 August 2024; positioned on par with [GPT-4o](https://aiwiki.ai/wiki/gpt_4o) [10] |
| GLM-4V | N/A | Multimodal (vision plus language) tier; see below [3] |

The exact roster and version dates have shifted over time as Zhipu added and retired snapshots; later additions such as GLM-4-FlashX sit alongside the original tiers on the platform.

## What is GLM-4-Plus?

GLM-4-Plus is the 2024 flagship of the GLM-4 line, announced by Zhipu on 29 August 2024 at the KDD 2024 conference [10]. It was released alongside a refreshed model suite: the text-to-image model CogView-3-Plus, the image and video understanding model GLM-4V-Plus, and the video generation model [CogVideoX](https://aiwiki.ai/wiki/cogvideo) [10]. Zhipu described GLM-4-Plus as delivering comprehensive improvements in language understanding, instruction following and long-text processing, positioning it as "on par" with first-tier models such as [GPT-4o](https://aiwiki.ai/wiki/gpt_4o), and a bilingual Chinese and English model with a 128K [context window](https://aiwiki.ai/wiki/context_window) and native tool use [10]. At the same event Zhipu launched what it described as the first video-call service for consumer users in China on its Qingyan app, spanning text, audio and video with real-time inference [10].

## GLM-4V multimodal

GLM-4V is the vision-language member of the family, accepting images alongside text. Zhipu positions it for document understanding, chart reading, optical character recognition, and general [visual question answering](https://aiwiki.ai/wiki/visual_question_answering) in both Chinese and English. An open 9-billion-parameter version, GLM-4V-9B, was released with the rest of the open series in June 2024 and operates at a native input resolution of 1120 by 1120 pixels with an 8K context window [8].

On Zhipu's reported multimodal evaluations, GLM-4V-9B is competitive with much larger proprietary systems on several benchmarks while trailing on others, notably the knowledge-heavy MMMU. Selected scores from the [model card](https://aiwiki.ai/wiki/model_card) are shown below [8].

| Benchmark | GLM-4V-9B | GPT-4o (0513) | GPT-4V (0409) | Gemini 1.0 Pro | Claude 3 Opus |
| --- | --- | --- | --- | --- | --- |
| MMBench-EN | 81.1 | 83.4 | 81.0 | 73.6 | 63.3 |
| MMBench-CN | 79.4 | 82.1 | 80.2 | 74.3 | 59.2 |
| SEEDBench-IMG | 76.8 | 77.1 | 73.0 | 70.7 | 64.0 |
| MMStar | 58.7 | 63.9 | 56.0 | 38.6 | 45.7 |
| MMMU | 47.2 | 69.2 | 61.7 | 49.0 | 54.9 |
| AI2D | 81.1 | 84.6 | 78.6 | 72.9 | 70.6 |
| OCRBench | 786 | 736 | 656 | 680 | 694 |

## Is GLM-4 open source?

Partly. The closed API tiers (GLM-4, GLM-4-Air, GLM-4-Flash, GLM-4-Plus and others) are commercial services billed per token on bigmodel.cn, with GLM-4-Flash offered free of charge. The open-weights branch is the GLM-4-9B series. On 5 June 2024 Zhipu open-sourced GLM-4-9B through the THUDM organization on [GitHub](https://aiwiki.ai/wiki/github) and Hugging Face [9]. The release includes a base model and several chat and multimodal variants:

| Model | Type | Context |
| --- | --- | --- |
| GLM-4-9B | Base | 8K |
| GLM-4-9B-Chat | Chat | 128K |
| GLM-4-9B-Chat-1M | Chat | 1M |
| GLM-4V-9B | Multimodal chat | 8K |

GLM-4-9B-Chat adds web browsing, code execution, [function calling](https://aiwiki.ai/wiki/function_calling) (custom tool use) and long-text reasoning on top of multi-turn dialogue. The 1M-context variant is positioned for whole-document workloads; Zhipu reports lossless retrieval in a "Needle in a Haystack" test at the 1M length [7][9].

On Zhipu's published comparisons, GLM-4-9B-Chat outperforms Llama-3-8B-Instruct across most general benchmarks. Reported chat-model scores are shown below [9].

| Model | MMLU | C-Eval | GSM8K | MATH | HumanEval | IFEval | MT-Bench |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Llama-3-8B-Instruct | 68.4 | 51.3 | 79.6 | 30.0 | 62.2 | 68.6 | 8.00 |
| GLM-4-9B-Chat | 72.4 | 75.6 | 79.6 | 50.6 | 71.8 | 69.0 | 8.35 |

The repository also reports base-model figures (for example, MMLU 74.7 and C-Eval 77.1 for GLM-4-9B before [instruction tuning](https://aiwiki.ai/wiki/instruction_tuning)) and multilingual results across six datasets where GLM-4-9B-Chat leads Llama-3-8B-Instruct [9].

## All-Tools agent capabilities

A central feature of the 2024 launch was "GLM-4 All Tools," an agentic mode in which the model decides on its own when and which tools to invoke rather than relying on a fixed pipeline [3]. The technical report describes four tool types: a web browser for retrieving and reading online information, a Python interpreter (code interpreter) for computation and data tasks, a text-to-image model, and user-defined functions [3]. The text-to-image component is CogView3, Zhipu's diffusion image generator [3]. The browser tool exposes operations such as searching and opening a page by URL, letting the model gather and synthesize current information across multiple steps.

The report states that in practical applications, GLM-4 All Tools "matches and even surpasses GPT-4 All Tools in tasks like accessing online information via web browsing and solving math problems using Python interpreter" [3]. The same All-Tools behavior, including web browsing, code execution and function calling, was carried into the open GLM-4-9B-Chat model [9].

## Long context

GLM-4 and the All Tools model both support a 128K-token context window, equivalent to roughly 300 pages of text [3][2]. Zhipu later pushed this further with two 1M-token offerings: the open GLM-4-9B-Chat-1M model and the GLM-4-Long API tier, the latter launched on 13 August 2024 and marketed for tasks such as translating long documents, analyzing financial reports and building [chatbot](https://aiwiki.ai/wiki/chatbot)s with very long memory [7]. The company has emphasized near-perfect recall at these lengths, reporting lossless [information retrieval](https://aiwiki.ai/wiki/information_retrieval) at 1M tokens in Needle-in-a-Haystack evaluations [7]. The context budget grew across generations of the GLM line, from a 2K window in the earliest ChatGLM models up to the 1M maximum reached here.

## How does GLM-4 compare to GPT-4 on benchmarks?

The technical report benchmarks the GLM-4 (0520) snapshot against GPT-4 (0314) and [GPT-4 Turbo](https://aiwiki.ai/wiki/gpt_4_turbo) (2024-04-09) on standard academic suites. GLM-4 closely rivals or beats the original GPT-4 on these general metrics while trailing the newer GPT-4 Turbo on several. Reported figures are below [3].

| Benchmark | GLM-4 (0520) | GPT-4 (0314) | GPT-4 Turbo (0409) |
| --- | --- | --- | --- |
| MMLU | 83.3 | 86.4 | 86.7 |
| GSM8K | 93.3 | 92.0 | 95.6 |
| MATH | 61.3 | 52.9 | 73.4 |
| BBH | 84.7 | 83.1 | 88.2 |
| GPQA | 39.9 | 35.7 | 49.3 |
| [HumanEval](https://aiwiki.ai/wiki/humaneval) | 78.5 | 67.0 | 88.2 |

On instruction following (IFEval, English strict-instruction), GLM-4 (0520) scores 85.0 against 85.9 for GPT-4 Turbo [3]. On the Chinese alignment benchmark AlignBench, GLM-4 (0520) and GPT-4 Turbo tie at 8.00 overall, and the report states GLM-4 "outperforms GPT-4 in Chinese alignments as measured by AlignBench" [3]. On the bilingual coding benchmark NaturalCodeBench, GLM-4 (0520) scores 47.1 overall versus 53.8 for GPT-4 Turbo [3]. These results are Zhipu's own measurements, so the precise figures depend on the prompts and evaluation settings used.

## Licensing

The open GLM-4-9B series is released under a custom Model License rather than a standard open-source license. The model weights are governed by the GLM-4 license posted on the Hugging Face repositories, while the accompanying code in the GitHub repository is released under Apache 2.0 [9]. The model license permits research and, subject to its terms, commercial use; users are directed to follow the license conditions, which historically have included a registration or notification requirement for commercial deployment. By contrast, the later GLM-4.5 and GLM-4.6 open weights moved to the permissive [MIT license](https://aiwiki.ai/wiki/mit_license), removing those restrictions [11][13].

## What came after GLM-4? GLM-4.5, GLM-4.6 and the Z.ai IPO

Zhipu continued to iterate on the GLM-4 line after the 2024 launch. In April 2025 it open-sourced the GLM-4-0414 series, including a 32-billion-parameter GLM-4-32B-0414 and the reasoning-focused GLM-Z1 and GLM-Z1-Rumination variants, all supporting context extension to 128K [9]. A separate audio model, [GLM-4-Voice](https://aiwiki.ai/wiki/glm_4_voice), added end-to-end speech conversation.

The family was then superseded by a new generation of Mixture-of-Experts (MoE) flagships built for agentic and coding workloads. GLM-4.5, released on 28 July 2025 under the MIT license, is a 355-billion-parameter MoE model with 32 billion active parameters (often written 355B-A32B), trained on 23 trillion tokens [11][12]. Z.ai reports that GLM-4.5 scores 63.2 averaged across 12 industry-standard benchmarks, third overall among the models tested, including 64.2 percent on [SWE-bench Verified](https://aiwiki.ai/wiki/swe-bench_verified), 91.0 percent on [AIME 2024](https://aiwiki.ai/wiki/aime_2024) and 70.1 percent on TAU-bench [11][12]. The smaller GLM-4.5-Air variant scores 59.8 on the same composite [11]. GLM-4.6, released in late September 2025, kept the 355B MoE design while extending the context window to 200K tokens and improving real-world coding and agentic search; it too is published under the MIT license [13][14].

The GLM line is the backbone of a company that became China's first publicly listed large-model developer. Zhipu AI rebranded its international identity as Z.ai in July 2025 (retaining the Zhipu AI / Knowledge Atlas name in China), and on 8 January 2026 it listed on the Hong Kong Stock Exchange, raising about HK$4.35 billion (around US$560 million) in what was described as the first IPO by one of China's "AI tiger" model startups [15][16].

## References

1. [Zhipu AI Unveils Next-Gen Foundation Model GLM-4, Claims Performance Comparable to GPT-4 (Maginative)](https://www.maginative.com/article/zhipu-ai-unveils-next-gen-foundation-model-glm-4-claims-performance-comparable-to-gpt-4/)
2. [Zhipu AI Introduces GLM-4 Model: Next-Generation Foundation Model Comparable with GPT-4 (MarkTechPost)](https://www.marktechpost.com/2024/01/23/zhipu-ai-introduces-glm-4-model-next-generation-foundation-model-comparable-with-gpt-4/)
3. [ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 (All Tools) (arXiv:2406.12793)](https://arxiv.org/abs/2406.12793)
4. [Chatbot Arena issue: Add GLM-4 variants AirX, Air and Flash (lm-sys/FastChat #3420)](https://github.com/lm-sys/FastChat/issues/3420)
5. [GLM-4 series repository README (zai-org/GLM-4 on GitHub)](https://github.com/zai-org/GLM-4)
6. [Zhipu AI: GLM-4-Flash Large Model API Interface Free to the Public (AIbase)](https://www.aibase.com/news/11312)
7. [Zhipu AI: GLM-4-Long API Launched with Input and Output Price of 0.001 Yuan / 1,000 Tokens (AIbase)](https://news.aibase.com/news/11031)
8. [THUDM/glm-4v-9b model card (Hugging Face)](https://huggingface.co/THUDM/glm-4v-9b/blob/main/README_en.md)
9. [GLM-4-9B open-source release notes, README_20240605 (zai-org/GLM-4 on GitHub)](https://github.com/zai-org/GLM-4/blob/main/README_20240605.md)
10. [Zhipu Releases Next-Generation Foundation Model GLM-4-Plus and Upgrades Video Call Feature of Qingyan APP (AIbase)](https://www.aibase.com/news/11387)
11. [Zhipu AI (Z.ai) Releases Open-Weights GLM-4.5 Models That Perform Comparably To the Latest from Claude and DeepSeek (DeepLearning.AI, The Batch)](https://www.deeplearning.ai/the-batch/zhipu-ai-z-ai-releases-open-weights-glm-4-5-models-that-perform-comparably-to-the-latest-from-claude-and-deepseek)
12. [GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models (arXiv:2508.06471)](https://arxiv.org/abs/2508.06471)
13. [zai-org/GLM-4.6 model card (Hugging Face)](https://huggingface.co/zai-org/GLM-4.6)
14. [Zhipu AI Releases GLM-4.6: Enhancements in Real-World Coding, Long-Context Processing, Reasoning, Searching and Agentic AI (MarkTechPost)](https://www.marktechpost.com/2025/09/30/zhipu-ai-releases-glm-4-6-achieving-enhancements-in-real-world-coding-long-context-processing-reasoning-searching-and-agentic-ai/)
15. [Z.ai (company overview) (Wikipedia)](https://en.wikipedia.org/wiki/Z.ai)
16. [The first of China's 'AI tigers' goes public as Zhipu climbs in Hong Kong debut (CNBC)](https://www.cnbc.com/2026/01/08/china-ai-tiger-goes-ipo-zhipu-hong-kong-debut-openai-knowledge-atlas-hsi-hang-seng-listing.html)