DeepSeek has released V4.1-Flash-1017, an update to its artificial intelligence model that the company says improves benchmark results while completing tasks faster and generating shorter responses. The model is available through DeepSeek’s website, mobile app and developer platform.
The Hangzhou-based company announced the release on Saturday, describing it as a minor update to V4.1-Flash rather than a new model generation. Its stated improvements span reasoning, coding, agentic tasks and visual benchmarks, with applications ranging from creating games to producing 3D models.
The release adds another iteration to DeepSeek’s competition with US developers including OpenAI and Anthropic, as well as Chinese rivals such as Moonshot Kimi and Z.ai. That competition extends beyond the quality of answers: developers also weigh the time a model takes to finish work, the volume of text it produces and the cost of running it repeatedly.
DeepSeek published benchmark results on X and described the update in a technical report hosted on Hugging Face. For Humanity’s Last Exam, it reported scores ranging from 36.8 to 42.6 for the updated model, compared with 36.8 to 39.1 for its predecessor. The available release details do not establish the evaluation conditions behind each figure, so those ranges should not be treated as a single, directly comparable result.
Humanity’s Last Exam tests models with challenging questions across mathematics, science and the humanities. Such evaluations provide one measure of capability, but they do not establish how reliably a model will perform in a particular application. Coding systems, for example, must also be assessed on whether their output runs correctly and fits the requirements of the software being developed.
The company’s claims of faster task completion and shorter outputs are particularly relevant to agentic workflows, in which a model may take several steps, use tools and revise its work before returning a result. In general, reducing unnecessary output can lower the amount of generated text a system must process. Whether that translates into lower costs or faster execution depends on pricing, the task and the surrounding software. The supplied release information does not quantify those savings for this update.
The launch follows the introduction of the slimmed-down V4.1-Flash in early September. DeepSeek has used aggressive pricing to compete for developers, placing pressure on other providers in a market where access costs can influence which models are selected for large-scale applications.
The update also arrives after a service outage earlier in the week that lasted about five hours and left DeepSeek’s servers unreachable. The disruption highlights a separate consideration from benchmark performance: users of hosted models depend on service availability as well as the underlying system’s capabilities. Although the new version is now live across DeepSeek’s consumer and developer channels, the release details do not establish whether it changes the infrastructure constraints affecting those services.
Sources: The Economic Times, Investing.com, NewsBytes, ProPakistani