The large model R1 of the open source of the Chinese DeepSeek team shows a remarkable advantage in terms of performance and cost, causing widespread attention in the global science and technology community. It surpassed the OPNAI O1 model in a number of authoritative testing, especially in the field of mathematics and programming, and stood out with a very low cost advantage, becoming a dark horse in the field of open source models. The open source of R1 not only demonstrates China's breakthrough in the field of big model technology, but also has injecting new vitality into global AI development.
Text: Recently, China's Deepseek team has launched its latest open source model R1, which has received widespread attention. The performance of the R1 model is extremely good, and it surpasses the OPENAI O1 model in many tests, especially in mathematical and programming evaluations.
In the latest American AIME2024 test, R1 surpassed 79.2 points in O1 with a score of 79.8. In the Math-500 test, R1 scored 97.3 points, which also led 96.4 points in O1. In addition, in the Swe-Bench Verify test, R1 scored 49.2, which also exceeded O1's 48.9 points. Although in code test Codeforcess, R1 is only 0.3 points lower than O1, the overall performance is equivalent to the O1 model.
In addition to performance, the cost advantage of R1 is more eye -catching. OPNAI's O1 model is as high as $ 15 per 1 million tokens, while the cost of R1 is only $ 0.14, a reduction of 90%. In terms of output, O1 costs $ 60 per million tokens, while R1 only costs $ 2.19, a decrease of 27 times. This huge cost difference makes R1 stand out in the field of open source models.
After the Deepseek team announced the open source of R1, many foreign netizens have expressed their admiration for this model, and believe that R1 has surpassed the old open source platforms such as Meta and Mistral in terms of cost performance and performance. Many people say that the efficient reasoning capabilities of the R1 model make it perform well in terms of code writing and mathematical interpretation. Some users even call it "the most like -white model of human beings." At the same time, Apple's machine learning researcher Awni Hannun also tested R1, and found that it was running rapidly and high reasoning efficiency on Apple M2ultra.
The development of the R1 model has gone through multiple stages of training processes, including cold startup data and multi -stage training to improve its reasoning ability and readability. The improvement of these technologies ensures the excellent performance of the R1 model in various tasks.
With the release of R1, China's open source model has once again attracted great attention and discussion in the international market. Many technical enthusiasts have expected the potential of this model. The release of R1 marks China's further breakthrough in the field of big model technology and has promoted the development of open source technology.
Open source address: https://huggingface.co/deepseek-ai/r1
API: https://api-docs.deepseek.com/guides/reasoning_model
Points:
The R1 model surpasses OPNAI O1 in a number of tests, showing excellent performance.
The input and output costs of R1 are as low as 0.14 US dollars and $ 2.19, respectively, with a decrease of 90%.
After the open source of R1, many foreign experts appreciate their performance and believe that its cost performance is very high.
The emergence of R1 not only provides developers with a high -performance, low -cost powerful tool, but also heralds the continuous innovation and competitiveness of China's continuous innovation and competitiveness in the field of artificial intelligence. It is expected that R1 will bring breakthrough progress to more fields in the future.