Search AI Products and News

On June 23, NetEase Youdao officially launched and open-sourced its latest 'Ziyue 3' series large model, named 'Confucius3-Math' in English. This reasoning model, focused on mathematical education, can efficiently run on ordinary consumer-grade GPUs, becoming the first low-cost and high-performance AI educational tool in China. In a series of mathematical reasoning tasks, 'Ziyue 3' has demonstrated outstanding performance that surpasses many large-scale general models. The 'Ziyue Math Model' achieves this through large-scale reinforcement learning and a series of innovative algorithms.

A recent joint study by Google, Carnegie Mellon University, and MultiOn explores the application of synthetic data in training large language models. According to Epoch AI, a research institution focused on AI development, currently available high-quality text training data totals around 300 trillion tokens. However, with the rapid advancement of large models like ChatGPT, the demand for training data is growing exponentially, projected to exhaust existing resources by 2026. Therefore, synthetic data is becoming increasingly crucial.

On March 6, 2025, a new language model called **Light-R1-32B** was officially launched. Built upon the **Qwen2.5-32B-Instruct** model, this mathematical problem-solving tool, specially trained for exceptional mathematical abilities, low training costs, and reproducibility, is a significant highlight in the field of artificial intelligence. The development team xAI stated that Light-R1-32B not only surpasses similar models in performance but also provides significant benefits for academic research and practical applications.

Today, the Alibaba Cloud Tongyi team officially released the new mathematical reasoning process reward model Qwen2.5-Math-PRM. This model offers two sizes, 72B and 7B, with performance significantly outperforming similar open-source process reward models, especially excelling in identifying reasoning errors. The 7B version of Qwen2.5-Math-PRM astonishingly surpasses the widely popular GPT-4o, marking an important step in Alibaba Cloud's research and development of reasoning models.

Today, iFLYTEK Spark 4.0 Turbo has undergone a comprehensive upgrade, with significant improvements in its seven core capabilities—text generation, language understanding, knowledge Q&A, logical reasoning, mathematical ability, coding ability, and multimodal ability—surpassing the powerful performance of GPT-4o in math and coding.

DeepSeek AI recently launched DeepSeek-V2.5-1210, an enhanced version of DeepSeek-V2.5, aimed at improving AI performance in math, programming, writing, and reasoning tasks. Earlier versions of the model achieved some success in solving math and reasoning tasks, but the stability of performance across various applications needed enhancement, especially in real-time coding and meticulous writing. These shortcomings underscore the potential for developing a more flexible and reliable AI model.

Recently, AI large language models (LLMs) have been performing remarkably well in various tasks, effortlessly writing poetry, coding, and chatting; they seem to be capable of anything! But can you believe it? These 'genius' AIs are actually 'math novices'! They often fail dramatically when tackling simple arithmetic problems, leaving people astonished. A recent study unveiled the 'quirky' secret behind LLMs' arithmetic reasoning abilities: they rely neither on powerful algorithms nor solely on memory, but instead adopt a strategy known as 'heuristic hodgepodge'!

The Dark Side of the Moon Kimi Smart Assistant has announced the launch of its next-generation mathematical reasoning model, k0-math. The k0-math model has shown outstanding performance in multiple mathematical benchmark tests, surpassing OpenAI's o1 series models, o1-mini and o1-preview, in four mathematical benchmark assessments including middle school entrance exams, college entrance exams, graduate school entrance exams, and MATH competitions that include introductory problems.

Following OpenAI's GPT-4 achieving remarkable results in traditional mathematics assessments, the research teams from Peking University and Alibaba have jointly launched a brand new evaluation benchmark - Omni-MATH, aimed at assessing the reasoning abilities of large language models at the level of the Olympic mathematics competitions. This initiative not only provides a new standard for evaluating AI mathematics capabilities but also opens up new avenues for exploring AI's potential in advanced mathematics. The unique design of Omni-MATH includes a database of 4,428 competition-level mathematics problems.

Recently, a Reddit user shared his experience on the forum using OpenAI's latest model, OpenAI o1, to solve high school math problems. Remarkably, he got all the answers correct. @FitAirline8359 mentioned using past math final exam questions from Chinese universities' upperclassmen and presented these problems to OpenAI's latest version, o1. To his surprise, OpenAI o1 answered all of them correctly in just 20 minutes. This result left him extremely shocked and sparked significant discussion among other users.

The Bilibili UP master 'Geek Bay' has created the 'Daydream' robot, which has quickly gone viral, surpassing 1.13 million views in just four days. This robot can automatically complete assignments, including flipping pages, framing questions, solving problems, and writing, covering multiple subjects. During the development process, the team progressed from modifying a writing machine to independent research and development, employing a CoreXY structure and an integrated AI model, significantly improving the accuracy and efficiency of homework completion. Test results show that the English homework has a high accuracy rate but formatting issues; the Chinese homework features unique handwriting, but some questions are incorrect; the math homework has formatting and logic problems.

Alibaba's Tongyi Qianwen team has released a math model named Qwen2 Math Demo, which excels at processing mathematical problems in text input and formulas in images and screenshots. It even surpasses GPT-4 on the MATH dataset, improving the score by 9.6%. The model has three versions, with the 72B version being particularly outstanding, outshining GPT-4 in challenging math problems and also outperforming Llama-3.1 on classic test sets like GSM8K and MATH.

Aliyun has launched the Qwen2-Math series of large language models, focusing on mathematical problem-solving abilities that surpass existing open-source models and rival well-known closed-source models, including GPT-4, Claude-3.5-Sonnet, Gemini-1.5-Pro, and Llama-3.1-405B. This series is based on the Qwen2 model, which has undergone deep pre-training on a specialized mathematical corpus that includes mathematical web texts, professional books, code examples, and exam questions, as well as data autonomously generated by Qwen2.