DeepSeek Makes V4-Pro 75% Discount Permanent — Further Fueling the Price War

DeepSeek has announced that the 75% discount on its flagship model, V4-Pro, will continue on a permanent basis. The discounted pricing, originally set to expire at the end of May, will simply be kept as is. Developer pricing remains at a quarter of the original rate.

According to Bloomberg, the move is expected to further intensify price competition across the AI industry. As Chinese AI companies compete more directly with their global rivals, the low-price strategy is a clear signal that DeepSeek is aiming for market penetration. US players such as OpenAI and Anthropic have already announced a series of API price cuts, and the downward trend in per-model pricing looks set to accelerate.

Alibaba Qwen3.7-Max Achieves 35 Hours of Autonomous Operation

Alibaba's Qwen team has released a new model, Qwen3.7-Max. A proprietary model designed for autonomous, long-running tasks, it delivered benchmark performance on par with Claude Opus 4.6 while outperforming domestic Chinese rivals such as DeepSeek V4 Pro and Kimi K2.6.

The most notable highlight was a demonstration in which the model autonomously performed code optimization for a custom chip over 35 hours. Completing long, complex tasks without human intervention marks a major milestone in AI agent autonomy. The team also released a quadruped robot control demo, signaling that applications in the physical world are in view.

What sets Qwen3.7-Max apart from earlier models is its specialization in long-running agent tasks. Rather than one-shot question answering, it targets use cases where work continues over periods on the order of tens of minutes to several hours.

UC Berkeley Law Bans AI from Nearly All Graded Assignments

UC Berkeley Law, one of the world's top law schools, has announced that starting with the summer 2026 term, AI will be banned in nearly all graded assignments. The ban covers a broad range of activities, from creating outlines and drafting text to proofreading. Only research use of AI will be permitted.

The school's rationale is clear: "Future lawyers must first develop the ability to think for themselves before they can use AI meaningfully." At the core of the policy is the judgment that building foundational legal reasoning skills—before reaching for AI tools—is central to legal education.

What makes the decision noteworthy is that, at a time when AI adoption is sweeping through education, the school has deliberately chosen to go against the current. It could also influence other educational institutions and corporate training programs.

Google CEO Pichai Calls Links "Part of Search" — Redefining the Web's Role?

Google CEO Sundar Pichai described links and sources in search results as "part of search." Even though links were once the very foundation of search, his phrasing—which seems to demote them to a mere "part"—is drawing attention.

According to The Decoder, the wording is deliberate. Google's new features show a growing tendency to keep users inside the Google ecosystem, and the company is shifting from "traffic distributor" to "AI publisher." Which sources get cited and which get left out is, in effect, becoming a question of "editorial power."

For website operators and the media industry—already facing a continued decline in traffic from Google Search—the remark only deepens their concerns.

LocalLLaMA Community Asks Whether "Peak Hype" Has Passed — Enthusiasm Cools

A debate is heating up in Reddit's LocalLLaMA community over whether AI has passed its "peak hype." Citing a slight dip in subscriber counts and Google Trends data, the poster argued that interest in AI is trending downward.

Some say this suggests a transition into the "trough of disillusionment" in Gartner's hype cycle, but others in the community push back with responses like "it's just a temporary correction" and "AI has simply entered the practical stage."

In reality, technical progress on local LLMs has not stopped—if anything, they are steadily reaching practical levels. There remains a strong view that the "quality" of interest is shifting from hype to practical use, and that this should not be judged by surface-level metrics alone.

---