The Rise and Fall of DeepSeek V4
DeepSeek, the Chinese AI company that once created a stir in the tech world with its “cost-effective” AI strategy, has faced a disappointing reception for its latest model, DeepSeek V4. After 15 months of development, the new model has not met expectations, particularly when compared to both U.S. and domestic competitors.
According to recent intelligence rankings released by Artificial Analysis on the 25th (local time), DeepSeek V4 Pro scored 52 points, placing it 10th overall. This score was significantly lower than OpenAI’s GPT-5.5, which earned 60 points and took the top spot. The gap between these models highlights the challenges DeepSeek faces in maintaining its competitive edge.
The situation became even more challenging for DeepSeek as it failed to outperform open-source models within China. Competitors like MoonShot AI’s Kimi K2.6 and Xiaomi’s MiMo V2.5 Pro both scored 54 points, placing them at 5th and 6th, respectively. This performance further tarnished DeepSeek’s reputation, as it no longer showcased the disruptive potential it once had.
Analysts noted that while DeepSeek V4 narrowly surpassed Geephu AI’s GLM 5.1, which scored 51 points, the model did not live up to the high expectations set by its previous releases.
Reactions from U.S. Experts
U.S. experts have expressed relief over the results of DeepSeek V4. Chris McGuire, a senior researcher at the Council on Foreign Relations (CFR), stated that V4 lacks competitiveness compared to America’s cutting-edge models and will not significantly narrow the gap in the AI field. Bloomberg reported that sighs of relief could be heard in Washington and Silicon Valley, indicating a sense of satisfaction among industry leaders.
Despite these challenges, DeepSeek V4 is not without its merits. According to DeepSeek, the model retains a context length of approximately 1 million characters—8 times longer than its predecessor. Additionally, it reduces KV cache usage, a metric for data processing efficiency, to 10% of previous levels. The company claims that these improvements have achieved world-class cost efficiency.
Key Features of DeepSeek V4
- Extended Context Length: The model can handle up to 1 million characters, making it suitable for complex tasks requiring extensive information processing.
- Improved Efficiency: KV cache usage has been reduced to 10% of previous levels, enhancing data processing efficiency.
- Cost-Effective Design: The model is designed to offer high performance at a lower cost, aligning with DeepSeek’s original strategy.
Future Prospects
While DeepSeek V4 may not have met all expectations, the company continues to innovate and refine its approach. The focus on cost efficiency and extended context length suggests that DeepSeek is still committed to its core principles. However, the challenge remains to close the gap with leading AI models from both the U.S. and other Chinese competitors.
As the AI landscape continues to evolve, DeepSeek will need to address the shortcomings highlighted by the recent rankings. The company’s ability to adapt and improve will be crucial in maintaining its position in the competitive AI market.
Membership Offers
For those interested in accessing premium content, there are several membership options available:
- Monthly Membership: 5900 won, offering access to a wide range of news and publications.
- Newspaper Readers: 2900 won, providing exclusive access to selected publications.
- Chosun Membership: Includes 55000 won worth of newspapers and magazines, along with 7000 points for convenient shopping.
These offers highlight the value of membership programs in providing access to quality content and additional benefits.
