DeepSeek has improved its model and worsened its political situation. The Chinese AI startup released DeepSeek-V4.1-Flash-1017 on October 17, 2026, promising better coding and faster completion of complex tasks as the U.S. government moved to remove DeepSeek technology from federal agencies.
The release includes the open-sourcing of the inference engine behind DeepSeek’s API. Its open-source models are widely available in countries including the United States and are mostly free to use, creating a strange split between broad public access and official U.S. restrictions.
DeepSeek’s own benchmark reports give V4.1-Flash-1017 a DeepSWE v1.1 software engineering score of 77.0, up from 74.2. The model also scored 42.3 on Terminal-Bench 4.0, while Artificial Analysis rated it at 45, compared with 39 for its predecessor.
The evaluation cost fell from $0.49 to $0.26 per task. That matters because the model’s pitch is not limited to raw capability; DeepSeek is presenting it as a cheaper engine for software work and complex tasks, where cost can decide whether developers use a model or leave it idle.
API pricing adds another layer. DeepSeek-V4.1-Flash-1017 costs $0.30 per 1 million uncached input tokens and $1.20 per million output tokens through its API at peak times, while the listed peak-hours rate is $1.50 per 1 million tokens and the off-peak rate is $0.75.
At $0.75 off-peak, the model ties with MiniMax-M3.1-Flash. The pricing is still a notable contrast with DeepSeek’s mostly free open-source models—because “open” is not the same thing as “free” once an API starts counting every token.
Washington’s ban targets federal systems
The U.S. Department of Homeland Security made its order public on October 19, 2026. DHS said the order was issued in late July and directs federal civilian agencies to exclude and remove all DeepSeek technology.
The order bans civilian agencies from procuring DeepSeek products and services. It includes agencies such as the U.S. Department of State, the U.S. Treasury Department, and the Internal Revenue Service, turning a model release into a procurement and security issue across the federal government.
The U.S. government alleges that DeepSeek’s models “present multiple threat vectors.” One concern is that DeepSeek’s servers are located in China, where Chinese authorities could access private company infrastructure and sensitive U.S. data.
That argument places DeepSeek inside the wider U.S.-China technology dispute rather than treating it as a normal software product. U.S.-led restrictions already ban China from accessing some advanced AI chips and the machines that make them, while Beijing has proposed new plans to restrict exports of certain advanced AI models.
The restrictions are expanding beyond models and chips. The U.S. Federal Communications Commission announced that it was barring imports of new foreign-made humanoid robots, adding another technology category to the growing list of products caught in national-security policy.
AI becomes another item on the Trump-Xi agenda
China’s Foreign Ministry spokesperson Mao Ning said China will safeguard the rights and interests of Chinese companies. Mao Ning also said, “China consistently opposes the overstretching of the concept of national security and the politicizing of science, technology, economic and trade issues.”
The dispute arrives before a scheduled meeting between U.S. President Donald Trump and Chinese President Xi Jinping in mid-November at the APEC summit in Shenzhen, China. AI is expected to be on the agenda, alongside the broader conflict over chips, exports, companies, and access to technology.
The economic backdrop is substantial. China’s AI-fueled IPO boom reached $54 billion in 2026, with that figure recorded on August 31, 2026, after a report on China’s AI IPO activity dated August 25.
DeepSeek’s release therefore lands in two opposing markets: one that rewards open models, low API costs, and stronger coding scores, and another that treats the same technology as a government security risk. V4.1-Flash-1017 may be faster and cheaper, but Washington has made clear that performance will not settle the argument.
Based on
- DeepSeek open-sources the inference engine behind its API. And Huawei gets equal billing with Nvidia. — thenewstack.io
- DeepSeek-V4.1-Flash-1017 arrives with improved benchmarks just as the U.S. government bans it from agencies — what enterprises should do | VentureBeat — venturebeat.com
- DeepSeek fires back at U.S.federal agency ban as ‘overstretching of the concept of national security’ | Fortune — fortune.com
