‘Faster and leaner’: DeepSeek rolls out upgraded AI model after recent service disruptions

San Francisco, October 19 - Chinese artificial intelligence startup DeepSeek has unveiled a faster and more efficient version of its AI model, DeepSeek-V4.1-Flash-1017, following recent service disruptions. According to Bloomberg, the updated model was made available on October 17 through the company's website, app, and developer platform.

The new DeepSeek model achieved a score of 42.6 on Humanity’s Last Exam without tools, an improvement from the previous V4.1-Flash version's 36.8 score. Published evaluations on Hugging Face also revealed enhancements in software development and other tasks involving AI agents.

DeepSeek claims that the update aims to decrease the time required to complete tasks and produce shorter responses, potentially lowering the cost of using the model. The company's technical report illustrated improvements, with the Terminal-Bench 4.0 score for coding agents increasing from 31.2 to 42.3 and DeepSWE v1.1 rising from 74.2 to 77.0. These figures are based on DeepSeek's own assessments, and do not conclusively determine the model's performance in all real-world applications.

The release of DeepSeek-V4.1-Flash-1017 comes amid intense competition among Chinese AI developers and major US firms like OpenAI and Anthropic, as well as local rivals, to create capable systems at lower costs. The September launch of the smaller V4.1-Flash model put pressure on competitors, causing shares of MiniMax Group, Z.ai, and Alibaba Group Holding to fall on the Hong Kong stock market.

The latest update follows recurring service slowdowns and an outage lasting approximately five hours earlier in the week, which left DeepSeek's servers inaccessible. Some experts suggested the disruption was due to a shortage of computing resources, as demand for DeepSeek's competitively priced AI models has strained its services.