DeepSeek has shipped another update to its flagship efficient model line. The Chinese AI lab unveiled V4.1-Flash-1017 on October 17 through a series of posts on X, describing it as a faster, leaner refresh of the earlier V4.1-Flash with “stronger benchmark performance, faster task completion, and shorter outputs than the previous version.” The model is live across DeepSeek’s website, its mobile app, and its developer platform.
According to the company’s technical report on Hugging Face, the update delivers substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks. DeepSeek disclosed benchmark numbers on X: on Humanity’s Last Exam, a broad test spanning math, science, and the humanities, the revised model scored between 36.8 and 42.6, up from 36.8 to 39.1 for the previous version. As with any vendor-reported score, independent verification is still pending.
The release keeps pressure on the economics of frontier AI. DeepSeek’s aggressively priced open-weight models, with MIT-licensed weights published on Hugging Face, continue to challenge closed offerings from ChatGPT, Claude, and Grok on cost, even as those rivals lead on raw capability. The lab’s earlier V4.1-Flash launch in September was sharp enough to move Hong Kong-listed AI stocks.
The launch comes days after DeepSeek suffered a roughly five-hour outage that left its servers unreachable, a reminder of the infrastructure strain behind surging developer demand. For a deeper look at low-cost options, see our guide to the best free AI tools in 2026, and explore the model itself on our DeepSeek tool page.
Source: The Straits Times
