The renowned AI model evaluation platform LmArena recently announced a major update, launching two new DeepSeek models named "a very secret and interesting model" and "a highly confidential and happy robot," which are highly entertaining names. This mysterious release has immediately sparked widespread attention and heated discussions within the AI community.
Although specific technical details about these two models are not yet fully disclosed, their unique naming style and DeepSeek's consistent track record of technological innovation are enough to ignite industry anticipation. This humorous and mysterious naming approach not only reflects DeepSeek's distinctive corporate culture but also suggests that these two models may have groundbreaking innovations in functionality or application scenarios.
As a leading Chinese AI research company, DeepSeek has rapidly gained prominence in the global AI field since its establishment in 2023, thanks to its open-source model strategy and efficient training technology. Its flagship models, DeepSeek-R1 and V3, have performed exceptionally well in multiple benchmark tests such as mathematics, programming, and general reasoning, with performance even comparable to top-tier models like OpenAI's o1 and Google's Gemini 2.5 Pro.
Notably, the DeepSeek-R1-0528 achieved a significant improvement in accuracy on the AIME2025 math test, rising from 70% to 87.5%, demonstrating its remarkable progress in complex reasoning tasks. The two new models released this time continue DeepSeek's tradition of innovation and are expected to further strengthen its market competitiveness in specific application areas.
LmArena, as an open and transparent AI model evaluation platform, is widely recognized for its reliability and fairness. The platform provides important references for developers to choose suitable models through real user interactions and practical task testing. Previously, the DeepSeek V3-0324 model had shown excellent performance in LmArena's math tests, surpassing strong competitors like Qwen and Gemini 2.5.
Although the specific functional specifications of "a very secret and interesting model" and "a highly confidential and happy robot" have not been officially released, their creative naming has already triggered widespread speculation within the community. Some analysts believe that "the interesting model" may be specifically optimized for creative writing or entertainment applications, while "the happy robot" could focus on providing more natural and friendly conversational experiences.
DeepSeek has always centered its development philosophy around an open-source strategy. Its models, such as R1 and V3, are released under the MIT license, allowing developers to freely modify and commercialize them. This open approach has enabled DeepSeek to establish a strong leading position in the open-source AI field.
More impressively, DeepSeek demonstrates exceptional cost control capabilities. It is reported that the training cost of its V3 model is approximately $6 million, far less than the $100 million training cost of GPT-4, making DeepSeek an extremely cost-effective AI solution.
However, recent reports indicate that DeepSeek has postponed the release plan for its R2 model due to chip supply limitations, which may pose certain challenges to its subsequent technological development. In this context, whether the new models can continue DeepSeek's successful trajectory remains to be determined through actual testing and application verification.
With the official launch of these two mysterious models on the LmArena platform, DeepSeek has undoubtedly reignited the innovation enthusiasm of the entire AI community. Although specific performance metrics and application details remain to be officially disclosed, the technological innovation potential behind them has already generated great expectations within the industry.
This release further reinforces the important role of open-source AI models in the global artificial intelligence ecosystem, showcasing the strong strength of Chinese AI enterprises in technological innovation and product development.
