Alibaba Cloud Releases TONGYI QIANWEN 2.5: Chinese Large Model Performance Leads the Way, Surpassing GPT-4 Turbo comprehensively

In the field of artificial intelligence, Alibaba Cloud has once again demonstrated its strong research and development capabilities. Recently, Alibaba Cloud officially released the 2.5 version of Tongyi Qianwen, a Chinese large-scale model that surpasses GPT-4 Turbo in terms of performance, and is hailed as the most powerful Chinese large-scale model in the world.

It is understood that the latest version of Tongyi Qianwen 2.5 has achieved significant results in multiple benchmark evaluations. Its latest open-source model with 110 billion parameters has surpassed Meta's Llama-3-70B model in MMLU, TheoremQA, GPQA, and other tests, becoming a new benchmark in the open-source field. This achievement not only proves the excellence of Tongyi Qianwen 2.5 in model parameters and performance, but also reflects Alibaba Cloud's leading position in the field of artificial intelligence.

Compared with the 2.1 version, Tongyi Qianwen 2.5 has made significant improvements in understanding ability, logical reasoning, instruction compliance, and code ability. Specifically, these abilities have improved by 9%, 16%, 19%, and 10% respectively, with Chinese language ability leading the industry. This leap forward has enabled Tongyi Qianwen 2.5 to achieve the same score as GPT-4 Turbo on the authoritative benchmark OpenCompass, making it the first domestic large-scale model to achieve such outstanding results in this benchmark.

In addition to the release of Tongyi Qianwen 2.5, Alibaba Cloud has also launched the latest open-source model Qwen1.5-110B. This model with 110 billion parameters has surpassed Meta's Llama-3-70B model in multiple benchmark evaluations and topped the Open LLM Leaderboard, an open-source large-scale model ranking list launched by HuggingFace. This achievement once again consolidates the leading position of the Tongyi open-source series in the industry.

In addition to the outstanding performance of the models, Tongyi's multimodal model and proprietary capability model have also demonstrated top-notch influence in the industry. Among them, the Tongyi Qianwen visual understanding model Qwen-VL-Max has surpassed Gemini Ultra and GPT-4V in multiple multimodal standard tests, and has been applied in many enterprises, bringing substantial help to various industries. In addition, the Tongyi Qianwen code large-scale model CodeQwen1.5-7B is also one of the leading models in the industry, ranking high on the HuggingFace code model list Big Code, and is also the foundation of Tongyi Lingma, the number one intelligent coding assistant in terms of user scale in China.

Alibaba Cloud stated that the release of Tongyi Qianwen 2.5 is one of the important achievements of Alibaba Cloud's continuous innovation in the field of artificial intelligence. In the future, Alibaba Cloud will continue to increase investment and research and development efforts in the field of artificial intelligence, and launch more high-performance and reliable models and products to provide enterprises and developers with higher quality and more efficient artificial intelligence services.