After OpenAI released a model with breakthrough reasoning capabilities, the competition for artificial intelligence reasoning capabilities has officially begun. Alibaba and Huansuan have successively launched heavyweight new products that not only match the performance of the o1 model but are also open source!On Thursday, Alibaba Tongyi Qianwen launched the QwQ-32B-Preview open-source model, which contains 32.5 billion parameters and can handle prompts of up to 32,000 tokens in length. In the AIME and MATH benchmark tests, it outperformed OpenAI's reasoning models o1-preview and o1-mini.QwQ is one of the few models that can compete with o1, excelling in mathematics and programming, especially in complex problems that require deep reasoning, and it can be used for commercial applications.Last week, the quantitative giant Huansuan's DeepSeek-R1-Lite model, in its Preview version, surpassed o1-preview in more challenging mathematical and coding tasks, significantly outpacing GPT-4o and others. In the AIME test benchmarks, its scores steadily improved as computation time increased.It is worth mentioning that the officials also stated that the model is still in the development stage and, after continuous iterations, the official version of the DeepSeek-R1 model will be fully open source.The emergence of Alibaba and Huansuan models indicates that reasoning AI is on the rise in the industry, which may provide opportunities for small AI developers to catch up and break the current dominance of a few tech giants.Fireworks, a startup that began researching reasoning models in the second quarter of this year, has its co-founder and CEO Lin Qiao stating:<blockquote>The entire open-source community... will launch reasoning models at an ultra-fast pace.</blockquote>In addition, tech giants have also increased their efforts in developing reasoning models. Google has expanded its reasoning model team from a few dozen people before the release of o1-preview to around 200 people, and Google has also provided more computational resources for this team.<h2>Latecomers Have Cost Advantages, Thinking Chains Are Key to Large Models</h2>Latecomers have a cost advantage in building large models.Latecomers seem to benefit from the reasoning papers published in recent years by researchers from Stanford University, Google, Meta Platforms, and OpenAI itself when developing alternatives to OpenAI. The development costs of reasoning models are lower than those of traditional LLMs, such as GPT-4o, which require hundreds of millions of dollars in computational resources and training data, and need to legally obtain this data.New models can help OpenAI and its competitors develop coding assistants capable of completing difficult projects. For example, enterprise software companies like Microsoft and Salesforce can use them to improve agents that take action on behalf of customers, such as scheduling appointments.It is worth mentioning that researchers can incorporate reasoning capabilities into existing LLMs by having other models generate the thought processes for solving problems and then using these processes to train LLMs.Some researchers have also made reasoning-focused datasets available for free to other developers. For example, Alibaba stated that it used data from one of OpenAI's research teams to build its reasoning modelIon Stoica, co-founder of AI startup Anyscale and Databricks, stated:<blockquote>In developing inference models, OpenAI's competitors do not have a clear disadvantage</blockquote>

OpenAI

BABA

The rise of inference large models in the industry provides an opportunity for small AI developers to catch up, and the development cost of inference models is lower than that of traditional large models. Latecomers can refer to research papers and data from OpenAI and others when building large models

- Following OpenAI's release of a groundbreaking reasoning model, a competition in AI reasoning capabilities has emerged, with Alibaba and Huanshuai launching significant new open-source models.  
- Alibaba's QwQ-32B-Preview model, featuring 32.5 billion parameters, outperforms OpenAI's models in AIME and MATH benchmarks, while Huanshuai's DeepSeek-R1-Lite excels in complex math and coding tasks.  
- This rise of new models indicates a shift in the AI landscape, potentially allowing smaller developers to compete against major tech firms, as the cost of developing reasoning models is lower than traditional large language models.

Wallstreetcn

Alibaba

Open source and capabilities on par with o1! Alibaba and Huanshou successively released heavyweight new products, with reasoning large models directly challenging OpenAI