Home Technology peripherals AI The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

Jun 10, 2024 am 11:44 AM
industry Wisdom spectrum ai GLM-4-9B

The latest version of the large model costs 6 cents and 1 million Tokens.

This morning, at the AI ​​Open Day, Zhipu AI, a large model company that has attracted much attention, announced a series of industry implementation figures:

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

According to the latest statistics, Zhipu AI large model open platform has currently obtained 300,000 registered users, and the average daily call volume has reached 40 billion Tokens, of which APIs have been used in the past 6 months. Daily consumption has increased by more than 50 times, and the most powerful GLM-4 model has increased by more than 90 times in the past 4 months.

In the recent Qingtan App, more than 300,000 agents have been active in the agent center, including many excellent productivity tools, such as mind maps, document assistants, schedulers, and more.

On the new technology side, the latest version of GLM-4, GLM-4-9B, surpasses Llama 3 8B in all aspects. The multi-modal model GLM-4V-9B is also online, and all large models remain open source. .

A series of commercial achievements and technological breakthroughs are eye-catching.

MaaS platform upgrade to version 2.0

Laying the threshold for large model application

Recently, domestic large models are setting off a new round competition.

In early May, Zhipu AI took the lead in reducing the price of the large model GLM-3-Turbo service to 1/5 of the original price, which also inspired many players in the large model field to "join the war." From the rush to establish start-up companies, the "Battle of 100 Models" to the price war, competition in the large model industry has spiraled upward.

Reducing the cost of large model services can allow more enterprises and developers to obtain new technologies, thereby generating sufficient usage. This will not only accelerate technological breakthroughs, but also allow large models to be used in various industries. Various industries are rapidly penetrating and commercializing the layout.

It is worth mentioning that at the current point, the price of large models has been pushed very low, but Zhipu said that it is not afraid of price war.

"I believe that everyone is familiar with the recent large-scale model price war, and is also very concerned about Zhipu's commercialization strategy. We are proud to say that through model core technology iteration and efficiency improvement, through technology Innovation enables continuous reduction of application costs while ensuring continuous upgrade of customer value," said Zhang Peng, CEO of Zhipu AI.

According to the different application scales of enterprises, Zhipu announced a series of latest price adjustments. The maximum API discount reaches 40% off, and the GLM-4-9B version can be used for only 6 cents / 1 million tokens. Looking back at the beginning of last year, the price of the large models of the GLM series has been reduced by 10,000 times.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

As the first startup to invest in generative AI, Zhipu AI’s commercialization speed is faster than that of many competitors. Build a product matrix based on hundreds of billions of multi-modal pre-trained models. It has launched a GLMs personalized agent customization tool for the C-side, allowing users to create their own GLM agents with simple prompt word instructions without any programming knowledge. For business-end customers, the latest generation of GLM-4 large models has been launched on the MaaS (Model as a Service) platform, providing API access.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

# This wisdom spectrum AI open platform.

On today’s Open Day, Zhipu launched the MaaS open platform 2.0, which has achieved improvements in new models, costs, security and other aspects.

At the event, Zhipu AI introduced the latest progress of its open platform. The upgraded model fine-tuning platform can help enterprises greatly simplify the process of building private models. The entire range of GLM-4 large models now supports deployment in just three steps.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

For technology implementation, model tools are only a small step. Zhipu CEO Zhang Peng has always believed that there are three model layers in large models, namely L0 (basic model), L1 (industry model) and L2 (inference model for segmented scenarios). This is a progressive relationship. What Zhipu has to do is to do its best to do L0, and then help its partners to do L1 and L2.

Zhipu AI’s commercialization path is based on the MaaS platform, and provides different solutions such as cloud API, cloud privatization, local privatization, and integrated hardware and software for different customer groups and needs. While meeting the needs of enterprises, it also achieves the scale of "models and services".

GLM-4 9B comprehensively surpasses Llama3

Multi-modal parity with GPT-4V, open source and free

###

For Zhipu AI, which regards building AGI as its goal, continuous iteration of large model technical capabilities is also a top priority.

Since the all In big model in 2020, Zhipu has been at the forefront of the artificial intelligence wave. Its research involves all aspects of large model technology, from the original pre-training framework GLM, domestic computing power adaptation, general base large models, to semantic reasoning, multi-modal generation, to long context, visual understanding, and Agent intelligence capabilities. In all aspects, Zhipu has invested considerable resources to promote original innovation in technology.

In the past year, Zhipu has successively launched four generations of general-purpose large models: ChatGLM in March 2023, ChatGLM2 in June, and ChatGLM3 in October last year; in January this year, the latest generation of base models Model GLM-4 officially released. At Open Day, Zhipu AI introduced to the outside world the latest open source achievement of the large base model GLM-4 - GLM-4-9B.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

#It is the open source version of the latest generation pre-training model GLM-4 series. GLM-4-9B has stronger basic capabilities, longer context, implements more precise function calls and All Tools capabilities, and has multi-modal capabilities for the first time.

Based on the powerful pre-training base, the comprehensive performance of GLM-4-9B in Chinese and English is 40% higher than that of ChatGLM3-6B, in Chinese alignment capability AlignBench, command compliance IFeval, engineering code Natural Code Bench, etc. Significant improvements have been achieved in benchmark data. Compared with Llama 3 8B, which has a larger amount of training, it is not inferior. It has a slight lead in English, and there is an improvement of up to 50% in Chinese subjects.

The context length of the new model has been extended from 128K to 1M, which means that the model can handle 2 million words of input at the same time, which is equivalent to two books of Dream of Red Mansions or 125 papers. On LongBench-Chat with a length of 128K, the GLM-4-9B-Chat model improves by 20% compared to the previous generation. In the needle-in-a-haystack test with a length of 1M, GLM-4-9B-Chat-1M also achieved a good result of all green.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

The new generation of large models also improves support for multiple languages. The model vocabulary has been upgraded from 60,000 to 150,000, and the coding efficiency of languages ​​other than Chinese and English has increased by an average of 30%, which means that the model can handle tasks in small languages ​​​​faster. Evaluations show that the ChatGLM-4-9B model’s multi-language capabilities comprehensively exceed Llama-3 8B.

While supporting local operation of consumer-grade graphics cards, GLM-4-9B not only demonstrates powerful dialogue capabilities, supports 1 million long texts, and covers multiple languages, and more importantly: intelligent The large models released by Pu are completely free and open source. Now, every developer can run this version of the GLM-4 model locally.

GitHub link: https://github.com/THUDM/GLM-4

Model: huggingface: https://huggingface.co/collections/THUDM/glm-4-665fcf188c414b03c2f7e3b7

Magic Community: https://modelscope.cn/organization/ZhipuAI

In addition to the powerful text model, Zhipu AI also open sourced the multi-modal model based on GLM-4-9B Model GLM-4V-9B. By adding Vision Transformer, this model achieves capabilities comparable to GPT-4V with only 13B parameters.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

#While technology evolves, the price of large models is also constantly decreasing. Zhipu has launched the GLM-4-AIR model, which basically retains the performance of the GLM-4 large model in January and has significantly reduced its price to 1 yuan/million tokens.

The performance of the GLM-4-Air is comparable to the larger GLM-4-0116 model at 1/100 the price. It is worth mentioning that the API of GLM-4-Air has greatly improved the inference speed. Compared with GLM-4-0116, the inference speed of GLM-4-Air has been increased by 200%, and it can output 71 tokens per second, which is far higher than that of GLM-4-0116. Faster than the reading speed of the human eye.

Zhipu stated that the price adjustment for large models is based on the comprehensive results of technological breakthroughs, computing power efficiency improvements and cost control. Price adjustments will be made at regular intervals in the future to better satisfy developers. , customer needs, highly competitive prices are not only reasonable, but also in line with their own business strategies.

Ecological construction enters the next level

As one of the first domestic startups to enter the large model track, Zhipu AI has now become the leader of domestic AI technology companies. represent.

It is not only the leader in domestic large model technology, but also a Chinese force that cannot be ignored in the large model academia and open source ecosystem. Zhipu has extensive influence in the field of AI, with cumulative downloads of open source models reaching 16 million times. Supporting the open source community is Zhipu’s unwavering commitment.

Going one step further, Zhipu AI is also jointly formulating AI safety standards for large models. On May 22, companies from different countries and regions, including OpenAI, Google, Microsoft and Zhipu AI, jointly signed the Frontier AI Safety Commitments. It points out that it is necessary to ensure a responsible governance structure and transparency for the safety of cutting-edge artificial intelligence, responsibly explain how to measure the risks of cutting-edge artificial intelligence models, and establish a clear process for risk mitigation mechanisms for cutting-edge artificial intelligence safety models.

Outside the field of AI, for many industries that have benefited from breakthroughs in large models, Zhipu AI is driving enterprise productivity changes through MaaS, and its large model ecosystem has begun to take shape. .

"Why do we judge that 2024 is the first year of AGI? If you can answer this question in one sentence: Scaling Law has not expired, and AI technology growth has entered a new stage. Large model technology innovation is still As the progress progresses by leaps and bounds, there are even signs that the speed is getting faster and faster," Zhang Peng said. "Frankly speaking, we have never seen a technology iteratively upgrade with such a steep innovation curve in history, and it lasts for such a long time."

Zhipu AI technology innovation and commercial implementation are speeding up Take this steep curve.

The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.

#In the process of technological development, Zhipu AI has been on the fast track.

The above is the detailed content of The open source version of GLM-4 is finally here: surpassing Llama3, multi-modality comparable to GPT4V, and the MaaS platform has also been greatly upgraded.. For more information, please follow other related articles on the PHP Chinese website!

Statement of this Website
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn

Hot AI Tools

Undresser.AI Undress

Undresser.AI Undress

AI-powered app for creating realistic nude photos

AI Clothes Remover

AI Clothes Remover

Online AI tool for removing clothes from photos.

Undress AI Tool

Undress AI Tool

Undress images for free

Clothoff.io

Clothoff.io

AI clothes remover

Video Face Swap

Video Face Swap

Swap faces in any video effortlessly with our completely free AI face swap tool!

Hot Tools

Notepad++7.3.1

Notepad++7.3.1

Easy-to-use and free code editor

SublimeText3 Chinese version

SublimeText3 Chinese version

Chinese version, very easy to use

Zend Studio 13.0.1

Zend Studio 13.0.1

Powerful PHP integrated development environment

Dreamweaver CS6

Dreamweaver CS6

Visual web development tools

SublimeText3 Mac version

SublimeText3 Mac version

God-level code editing software (SublimeText3)

DeepMind robot plays table tennis, and its forehand and backhand slip into the air, completely defeating human beginners DeepMind robot plays table tennis, and its forehand and backhand slip into the air, completely defeating human beginners Aug 09, 2024 pm 04:01 PM

But maybe he can’t defeat the old man in the park? The Paris Olympic Games are in full swing, and table tennis has attracted much attention. At the same time, robots have also made new breakthroughs in playing table tennis. Just now, DeepMind proposed the first learning robot agent that can reach the level of human amateur players in competitive table tennis. Paper address: https://arxiv.org/pdf/2408.03906 How good is the DeepMind robot at playing table tennis? Probably on par with human amateur players: both forehand and backhand: the opponent uses a variety of playing styles, and the robot can also withstand: receiving serves with different spins: However, the intensity of the game does not seem to be as intense as the old man in the park. For robots, table tennis

The first mechanical claw! Yuanluobao appeared at the 2024 World Robot Conference and released the first chess robot that can enter the home The first mechanical claw! Yuanluobao appeared at the 2024 World Robot Conference and released the first chess robot that can enter the home Aug 21, 2024 pm 07:33 PM

On August 21, the 2024 World Robot Conference was grandly held in Beijing. SenseTime's home robot brand "Yuanluobot SenseRobot" has unveiled its entire family of products, and recently released the Yuanluobot AI chess-playing robot - Chess Professional Edition (hereinafter referred to as "Yuanluobot SenseRobot"), becoming the world's first A chess robot for the home. As the third chess-playing robot product of Yuanluobo, the new Guoxiang robot has undergone a large number of special technical upgrades and innovations in AI and engineering machinery. For the first time, it has realized the ability to pick up three-dimensional chess pieces through mechanical claws on a home robot, and perform human-machine Functions such as chess playing, everyone playing chess, notation review, etc.

Claude has become lazy too! Netizen: Learn to give yourself a holiday Claude has become lazy too! Netizen: Learn to give yourself a holiday Sep 02, 2024 pm 01:56 PM

The start of school is about to begin, and it’s not just the students who are about to start the new semester who should take care of themselves, but also the large AI models. Some time ago, Reddit was filled with netizens complaining that Claude was getting lazy. "Its level has dropped a lot, it often pauses, and even the output becomes very short. In the first week of release, it could translate a full 4-page document at once, but now it can't even output half a page!" https:// www.reddit.com/r/ClaudeAI/comments/1by8rw8/something_just_feels_wrong_with_claude_in_the/ in a post titled "Totally disappointed with Claude", full of

At the World Robot Conference, this domestic robot carrying 'the hope of future elderly care' was surrounded At the World Robot Conference, this domestic robot carrying 'the hope of future elderly care' was surrounded Aug 22, 2024 pm 10:35 PM

At the World Robot Conference being held in Beijing, the display of humanoid robots has become the absolute focus of the scene. At the Stardust Intelligent booth, the AI ​​robot assistant S1 performed three major performances of dulcimer, martial arts, and calligraphy in one exhibition area, capable of both literary and martial arts. , attracted a large number of professional audiences and media. The elegant playing on the elastic strings allows the S1 to demonstrate fine operation and absolute control with speed, strength and precision. CCTV News conducted a special report on the imitation learning and intelligent control behind "Calligraphy". Company founder Lai Jie explained that behind the silky movements, the hardware side pursues the best force control and the most human-like body indicators (speed, load) etc.), but on the AI ​​side, the real movement data of people is collected, allowing the robot to become stronger when it encounters a strong situation and learn to evolve quickly. And agile

ACL 2024 Awards Announced: One of the Best Papers on Oracle Deciphering by HuaTech, GloVe Time Test Award ACL 2024 Awards Announced: One of the Best Papers on Oracle Deciphering by HuaTech, GloVe Time Test Award Aug 15, 2024 pm 04:37 PM

At this ACL conference, contributors have gained a lot. The six-day ACL2024 is being held in Bangkok, Thailand. ACL is the top international conference in the field of computational linguistics and natural language processing. It is organized by the International Association for Computational Linguistics and is held annually. ACL has always ranked first in academic influence in the field of NLP, and it is also a CCF-A recommended conference. This year's ACL conference is the 62nd and has received more than 400 cutting-edge works in the field of NLP. Yesterday afternoon, the conference announced the best paper and other awards. This time, there are 7 Best Paper Awards (two unpublished), 1 Best Theme Paper Award, and 35 Outstanding Paper Awards. The conference also awarded 3 Resource Paper Awards (ResourceAward) and Social Impact Award (

Li Feifei's team proposed ReKep to give robots spatial intelligence and integrate GPT-4o Li Feifei's team proposed ReKep to give robots spatial intelligence and integrate GPT-4o Sep 03, 2024 pm 05:18 PM

Deep integration of vision and robot learning. When two robot hands work together smoothly to fold clothes, pour tea, and pack shoes, coupled with the 1X humanoid robot NEO that has been making headlines recently, you may have a feeling: we seem to be entering the age of robots. In fact, these silky movements are the product of advanced robotic technology + exquisite frame design + multi-modal large models. We know that useful robots often require complex and exquisite interactions with the environment, and the environment can be represented as constraints in the spatial and temporal domains. For example, if you want a robot to pour tea, the robot first needs to grasp the handle of the teapot and keep it upright without spilling the tea, then move it smoothly until the mouth of the pot is aligned with the mouth of the cup, and then tilt the teapot at a certain angle. . this

Hongmeng Smart Travel S9 and full-scenario new product launch conference, a number of blockbuster new products were released together Hongmeng Smart Travel S9 and full-scenario new product launch conference, a number of blockbuster new products were released together Aug 08, 2024 am 07:02 AM

This afternoon, Hongmeng Zhixing officially welcomed new brands and new cars. On August 6, Huawei held the Hongmeng Smart Xingxing S9 and Huawei full-scenario new product launch conference, bringing the panoramic smart flagship sedan Xiangjie S9, the new M7Pro and Huawei novaFlip, MatePad Pro 12.2 inches, the new MatePad Air, Huawei Bisheng With many new all-scenario smart products including the laser printer X1 series, FreeBuds6i, WATCHFIT3 and smart screen S5Pro, from smart travel, smart office to smart wear, Huawei continues to build a full-scenario smart ecosystem to bring consumers a smart experience of the Internet of Everything. Hongmeng Zhixing: In-depth empowerment to promote the upgrading of the smart car industry Huawei joins hands with Chinese automotive industry partners to provide

Distributed Artificial Intelligence Conference DAI 2024 Call for Papers: Agent Day, Richard Sutton, the father of reinforcement learning, will attend! Yan Shuicheng, Sergey Levine and DeepMind scientists will give keynote speeches Distributed Artificial Intelligence Conference DAI 2024 Call for Papers: Agent Day, Richard Sutton, the father of reinforcement learning, will attend! Yan Shuicheng, Sergey Levine and DeepMind scientists will give keynote speeches Aug 22, 2024 pm 08:02 PM

Conference Introduction With the rapid development of science and technology, artificial intelligence has become an important force in promoting social progress. In this era, we are fortunate to witness and participate in the innovation and application of Distributed Artificial Intelligence (DAI). Distributed artificial intelligence is an important branch of the field of artificial intelligence, which has attracted more and more attention in recent years. Agents based on large language models (LLM) have suddenly emerged. By combining the powerful language understanding and generation capabilities of large models, they have shown great potential in natural language interaction, knowledge reasoning, task planning, etc. AIAgent is taking over the big language model and has become a hot topic in the current AI circle. Au

See all articles