DeepSeek V4.1 Flash Launches and Surpasses V4 Pro Across the Board; OpenAI Pauses Pro Subscriptions

Leading the Pack: DeepSeek V4.1 Flash Outperforms V4 Pro on Every Front

The AI industry is seeing a flurry of updates. According to IT之家, DeepSeek today officially launched the V4.1 Flash model, which surpasses its predecessor, the flagship V4 Pro, across four dimensions: performance, cost, speed, and total response time. This means the Flash series is no longer just a cheap, lightweight option — it has genuinely overtaken the premium tier in overall experience.

Notably, DeepSeek has adopted a seamless migration strategy: existing API calls directed at V4 Pro will automatically switch to the new model, allowing developers to enjoy the upgrade without changing a single line of code. This "transparent upgrade" lowers the barrier to migration and reflects the vendor's confidence in the new model's stability.

On pricing, V4.1 Flash cuts the cache hit price to ¥0.02 per million tokens — a reduction of 60%. For applications with high call volumes and frequent repeated requests, this optimization translates to a dramatic drop in inference costs. When model capability and cost improve in tandem, small and mid-sized developers and early-stage startups gain significantly more room to deploy.

Compute Crunch: OpenAI Pauses New ChatGPT Pro Subscriptions

According to Sina Tech, OpenAI is considering pausing new ChatGPT Pro subscriptions due to unprecedented demand. The company stated that demand for Astra — its suite of advanced features — far exceeds current supply, and that it may close the Pro subscription channel to prioritize the experience of existing users and prevent compute resources from being overwhelmed by new traffic.

Interestingly, the last time OpenAI took such an extreme measure was during the GPT-4 Turbo era in 2023. The reemergence of a supply shortage years later reflects two converging trends: the compute requirements of top-tier models continue to climb, and the growth rate of premium paid users has outpaced infrastructure expansion.

Compute has become the central bottleneck constraining the scalability of large model services — which explains why major players are racing to invest in custom silicon and proprietary acceleration solutions. When "can't get enough GPUs" or "insufficient power" become real constraints, whoever breaks through the compute ceiling first will hold the upper hand in the next round of competition.

AI Safety Controversy: Researcher Resigns, Warns of "Betting Human Lives"

As the industry charges ahead, safety concerns are surfacing. According to Sina Finance, Jacob Coxon, a researcher who previously worked at both OpenAI and Anthropic, publicly resigned and stated bluntly that both companies are racing to develop self-improving superintelligence. He warned that this "race-to-the-top" development approach is effectively gambling with the lives of all humanity.

This kind of insider statement is not an isolated incident. In recent years, a growing number of AI safety researchers have left their positions and gone public with concerns about the pace of the industry's development. The core tension is clear: commercial competitive pressure drives companies to accelerate constantly, while safety evaluation and alignment research require far more careful, time-consuming processes. When the two come into conflict, safety issues are easily subordinated to growth targets.

This makes the path of policy regulation all the more critical going forward.

Policy and Law: AI Regulatory Frameworks Take Shape

OpenAI Supports California AI Legislation

In an interesting juxtaposition, OpenAI is simultaneously expanding and calling for oversight. According to Global Market Broadcast, OpenAI has expressed support for four AI-related bills in California and is calling for a mandatory national AI safety regulatory framework. Until Congress formally legislates, OpenAI says it will continue to support state-level AI lawmaking. The bills cover areas including independent evaluation, audit standards, and the protection of minors.

Mihoyo AI Voice Impersonation Case: ¥750,000 in Damages Awarded

Back home in China, the legal boundaries of AI are also being steadily defined. According to IT之家, Shanghai's first unfair competition case involving AI voice impersonation has reached a verdict: the voice of a Genshin Impact character was impersonated using AI and used for commercial gain, and the court ordered ¥750,000 in damages. The ruling sets a precedent for protecting "voice rights" in the AI era and sends a clear signal to the market — profiting from AI-generated impersonations of another person's voice carries legal liability.

Chip Wars: U.S. DOJ Investigates NVIDIA's Licensing Agreement with Grok

According to reports, the U.S. Department of Justice is investigating a licensing agreement between NVIDIA and Grok, examining whether NVIDIA used the licensing deal to sidestep antitrust scrutiny. Under the agreement, Grok licensed NVIDIA to use chips purpose-built for AI — a transaction that was accompanied by the departure of Grok's CEO and two senior executives.

Against the backdrop of AI chip shortages, leading companies are using licensing deals, acquisitions, and partnerships to consolidate resources across the supply chain — moves that can easily trigger regulatory concerns about market monopolization. How this investigation unfolds could shape the future landscape of partnerships and competition in the AI chip space.

Consumer AI: Apple Launches New Siri AI with Support for 16 Languages

Consumer-facing AI is also seeing new developments. According to IT之家, Apple unveiled a new Siri AI at its fall event, with plans to support 16 languages including Simplified Chinese — generating considerable buzz among users in mainland China.

Once a pioneer in voice assistants, Siri has been overtaken in terms of intelligence by newer entrants in recent years. With this rebuild powered by large model capabilities, Apple is attempting to reclaim consumer mindshare. Whether it can truly deliver on the promise of "intelligence" remains to be seen through real-world use, but multi-language support is undeniably a critical step toward addressing a global audience.

Closing Thoughts

From DeepSeek V4.1 Flash's dual breakthrough in performance and price, to OpenAI pausing subscriptions due to compute constraints; from a researcher's public warning about safety, to the steady advance of regulation and jurisprudence around the world — today's AI industry is marked by the parallel tracks of technological acceleration and institutional catch-up. Large models and compute are still leading the charge, but how to find the balance between speed and safety is the question the entire industry must now answer.