eWEEK content material and product suggestions are editorially unbiased. We could generate income if you click on on hyperlinks to our companions. Learn More.
Zhipu AI, a Chinese AI startup based in 2019, just lately launched its free AI agent to most people. Known as AutoGLM Rumination, the brand new answer, which has already secured tens of millions of {dollars} in government-backed funding, is making headlines throughout the business.
What is AutoGLM Rumination?
AutoGLM Rumination is likely one of the latest AI brokers to hit the buyer market. It’s able to performing fundamental net searches in addition to extra superior analysis duties. Current makes use of for AutoGLM Rumination embody technical writing and journey planning.
The AI agent is powered by two of Zhipu AI’s proprietary giant language fashions (LLMs): GLM-4-Air-0414 and GLM-Z1-Air. To assess how these LLMs measure up towards competing fashions, it’s important to look at accessible benchmark information.
Sizing up the competitors by means of benchmarks
Though solely in operation for a number of years, Zhipu AI builders have made bold claims relating to the efficiency of their generative AI instruments. Developers declare that GLM-Z1-Air performs eight instances sooner than DeepSeek-R1; it reportedly does so whereas solely utilizing a fraction of the computational energy.
A analysis paper printed in June 2024 reveals that Zhipu AI’s most up-to-date LLM, GLM-4, does surpass OpenAI’s GPT-Four throughout quite a few benchmarks. The paper’s authors said that GLM-4 “closely rivals or outperforms GPT-4 in terms of general metrics such as MMLU, GSM8K, MATH, BBH, GPQA, and HumanEval.”
However, it falls quick when in comparison with different sorts of AI fashions, corresponding to Claude 2, Claude 3 Opus, Gemini 1.5 Pro, and GPT-4 Turbo, in sure areas. In Python and Java programming, for instance, GLM-Four is the one Zhipu AI mannequin that scores excessive sufficient to offer any actual competitors — and even that lags behind the highest fashions on NaturalCodeBench.
While GLM-Four is main lots of the benchmarks on LongBench-Chat, GLM-4-Air and GLM-4-9B-Chat didn’t carry out fairly as nicely. They each struggled to maintain tempo with the competitors within the English language benchmarks, however all three carried out nicely with the Chinese language assessments.
Contributing to the open-source neighborhood
Zhipu AI has made quite a few contributions to the open-source AI neighborhood, and so they’ve gathered greater than 10 million downloads for his or her previous releases. These embody:
- ChatGLM-6B
- GLM-4-9B
- GLM-4V-9B
- WebGLM
- CodeGeeX
Despite a slew of AI startups coming into the market with their very own options, builders with Zhipu AI are already making their presence identified. AutoGLM Rumination is simply the most recent in a line of AI-driven merchandise, and we’ll possible hear extra from Zhipu AI within the close to future.
