Chinese AI startup DeepSeek released preview versions of its new flagship open-source model, V4, on Friday. The company launched two versions: DeepSeek-V4-Pro with 1.6 trillion parameters and DeepSeek-V4-Flash with 284 billion parameters. Both models feature a 1 million token context window and currently process text only. DeepSeek claims V4-Pro-Match achieved top-tier performance in coding benchmarks and matches OpenAI's GPT-5.4 on some tests. The company also says the V4 Pro Max version outperforms OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro. Some analysts note V4 is competitive with U.S. rivals but not as big a breakthrough as DeepSeek's earlier R1 model.
DeepSeek's V4 release intensifies the AI rivalry between China and the U.S. The new model excels at writing computer code, significantly outperforming every other open-source system in code generation tests from Vals AI. This follows DeepSeek's earlier R1 model, which showed an open-source system could perform almost as well as closed versions from U.S. companies. The release comes just days after another Chinese startup, Moonshot AI, introduced its latest open-source model, Kimi 2.6. Huawei announced full support of its Ascend chips for V4 models, and AI chipmaker Cambricon Technologies also announced compatibility.
In other AI news, South Korean police arrested a 40-year-old man for sharing an AI-generated image of a wolf named Neukgu that escaped from a zoo in Daejeon city. The photo misled authorities, prompting them to relocate their search and issue an emergency text to residents. The man told police he created the image for fun and faces up to five years in prison or a fine of 10 million Korean won for disrupting government work by deception.
Tennessee Tech University computer science professor William Eberle has been selected for the National Science Foundation's inaugural cohort of National AI Education Fellows. Eberle is among 25 faculty leaders from 23 states chosen to serve as national ambassadors for AI development. His selection comes as Tennessee Tech plans to launch a new bachelor's degree in artificial intelligence this fall.
Grow Therapy topped The Americas' Fastest-Growing Companies 2026 list with a compound annual growth rate of 455.6 percent. The company connects people with therapists and partners with more than 125 health insurers. Grow Therapy is testing AI's potential to improve mental health care. Its revenues grew from $3.6 million in 2021 to $617.4 million in 2024.
Serial entrepreneur Bill Nguyen is building an AI system that can operate as his proxy by absorbing information about how he communicates and what he values. The AI has permission to email people on his behalf and sometimes sets up in-person meetings. Nguyen pays per token and runs multiple models repeatedly to achieve higher levels of reliability, calling this approach agentic scaling.
AI is facing a growing backlash from the American public due to its energy consumption and impact on jobs. The article discusses whether Maine will become the first US state to adopt a moratorium suspending the construction of new data centers. Meanwhile, 85 percent of enterprises are running AI agent pilots but only 5 percent have moved those agents into production, according to Cisco's Jeetu Patel. The problem is the absence of a trust architecture for AI agents. Cisco launched Defense Claw at RSAC 2026 to address the trust deficit.
The AI compute crunch is affecting the labor market, gadget market, and electricity prices. AI companies are restricting access to their products, shutting down products, and beginning to increase prices. RAM, graphics cards, and storage for consumers have skyrocketed in price and are sold out in many stores. The age of cheap AI appears to be ending as venture capitalists and investment firms face higher costs to subsidize AI products.
AI expert Matt Pocock shared insights on effective LLM planning at an AI Engineer Europe talk. He introduced the concept of a smart zone and dumb zone for LLMs where the model performs best with clear concise prompts. Pocock advocated for a multi-phase plan approach that divides large tasks into smaller sequential steps. He also demonstrated his grill-me skill which helps developers stress-test an AI's understanding of a plan.
Apprenticeship.com announced its new job board for apprenticeships and paid work-based training opportunities. The platform connects job seekers with thousands of apprenticeship programs in skilled trades, healthcare, technology, and business. The launch comes as artificial intelligence reshapes entry-level work and accelerates demand for skills-based hiring. The platform is free to use for career seekers.
Key Takeaways
- DeepSeek released preview versions of its new flagship open-source AI model V4, with two versions: V4-Pro (1.6 trillion parameters) and V4-Flash (284 billion parameters).
- DeepSeek claims V4-Pro-Match matches OpenAI's GPT-5.4 on some tests and outperforms GPT-5.2 and Gemini 3.0-Pro.
- DeepSeek's V4 excels at coding, significantly outperforming all other open-source systems in code generation tests from Vals AI.
- Huawei and Cambricon Technologies announced compatibility with DeepSeek's V4 models.
- A South Korean man was arrested for sharing an AI-generated image of a runaway wolf, facing up to five years in prison.
- Tennessee Tech professor William Eberle was named a National AI Education Fellow by the NSF, as the university plans a new AI bachelor's degree.
- Grow Therapy topped the fastest-growing companies list with 455.6% CAGR, growing revenue from $3.6 million in 2021 to $617.4 million in 2024.
- Bill Nguyen is building an AI proxy that emails and schedules meetings on his behalf, using agentic scaling.
- 85% of enterprises run AI agent pilots, but only 5% have moved to production due to a lack of trust architecture; Cisco launched Defense Claw to address this.
- The AI compute crunch is driving up prices for RAM, graphics cards, and storage, and AI companies are restricting access and raising prices.
DeepSeek releases preview of new V4 AI model
Chinese AI startup DeepSeek released a preview of its new V4 large language model on Friday. The open-source model comes in pro and flash versions and shows strong performance in agent-based tasks and knowledge processing. DeepSeek claims V4 offers lower inference costs than previous models. The release comes as competition in China's AI sector intensifies following the success of DeepSeek's earlier R1 model.
DeepSeek launches new flagship open source AI model V4
DeepSeek released preview versions of its new flagship open source AI model V4 on Friday. The company launched two versions: DeepSeek-V4-Pro with 1.6 trillion parameters and DeepSeek-V4-Flash with 284 billion parameters. Both models have a 1 million token context window and can only process text for now. DeepSeek said V4-Pro-Match achieved top-tier performance in coding benchmarks and matches OpenAI's GPT-5.4 on some tests.
DeepSeek launches long-awaited V4 AI model update
DeepSeek launched preview versions of its latest major AI update on Friday as the AI rivalry between China and the U.S. heats up. The new V4 open-source models include pro and flash versions with improvements in knowledge, reasoning, and agentic capabilities. DeepSeek said the V4 Pro Max version has superior performance compared to OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro. Some analysts say V4 is competitive with U.S. rivals but not as big a breakthrough as the earlier R1 model.
DeepSeek V4 extends China's reach in open-source AI
Chinese startup DeepSeek released a preview of its V4 model on Friday, which it intends to open source. The new model excels at writing computer code and significantly outperformed every other open-source system at generating code according to tests from Vals AI. DeepSeek's earlier R1 model demonstrated that an open-source system could perform almost as well as closed versions from U.S. companies. The release comes just days after another Chinese startup Moonshot AI introduced its latest open-source model Kimi 2.6.
DeepSeek releases next-gen AI model with world-leading efficiency
DeepSeek released its next-generation foundational AI model V4 on Friday. The V4-pro version has 1.6 trillion parameters making it the company's biggest model ever while the V4-flash version has 284 billion parameters. Both models have a 1 million token context window achieved with world-leading cost efficiency. Huawei announced full support of its Ascend chips for V4 models and AI chipmaker Cambricon Technologies also announced compatibility with the new models.
DeepSeek rolls out long-anticipated AI model update
DeepSeek launched preview versions of its latest major AI update on Friday as the AI rivalry between China and the U.S. heats up. The new V4 open-source models include pro and flash versions with improvements in knowledge, reasoning, and agentic capabilities. DeepSeek said the V4 Pro Max version has superior performance compared to OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro. Some analysts say V4 is competitive with U.S. rivals but not as big a breakthrough as the earlier R1 model.
South Korea police arrest man for AI image of runaway wolf
South Korean police arrested a 40-year-old man for sharing an AI-generated image that misled authorities searching for a wolf named Neukgu that escaped from a zoo in Daejeon city. The photo prompted authorities to urgently relocate their search operation and issue an emergency text to residents. The man told police he created the image for fun. He faces up to five years in prison or a fine of 10 million Korean won for disrupting government work by deception.
Man faces prison for using AI to fake wolf sighting
A 40-year-old man was arrested in South Korea for using artificial intelligence to generate a fake image of a runaway wolf named Neukgu. The wolf had escaped from a zoo in Daejeon city and authorities launched an all-out effort to bring him back. The AI-generated photo prompted the city government to issue an emergency text warning residents of a wolf in the area. The man now faces a maximum sentence of five years in prison or a fine of up to $6,700.
Tennessee Tech professor named national AI education ambassador
Tennessee Tech University computer science professor William Eberle has been selected for the National Science Foundation's inaugural cohort of National AI Education Fellows. Eberle is among 25 faculty leaders from 23 states chosen to serve as national ambassadors for AI development. He brings nearly four decades of academic and industry experience to the fellowship. Eberle's selection comes as Tennessee Tech plans to launch a new bachelor's degree in artificial intelligence this fall.
Grow Therapy tops fastest-growing companies list with AI
Grow Therapy topped The Americas' Fastest-Growing Companies 2026 list with a compound annual growth rate of 455.6 percent. The company connects people with therapists and partners with more than 125 health insurers. Grow Therapy is testing AI's potential to improve mental health care. The company's revenues grew from $3.6 million in 2021 to $617.4 million in 2024.
Bill Nguyen builds AI assistant to act as his proxy
Serial entrepreneur Bill Nguyen is building an AI system that can operate as his proxy by absorbing information about how he communicates and what he values. The AI has permission to email people on his behalf and sometimes sets up in-person meetings. Nguyen pays per token and runs multiple models repeatedly to achieve higher levels of reliability. He calls this approach agentic scaling and says he paid to get a peek into the future of AI.
AI faces growing backlash from American public
AI is facing a growing backlash from the American public due to its energy consumption and impact on jobs. The article discusses whether Maine will become the first US state to adopt a moratorium suspending the construction of new data centers. The backlash reflects concerns about the environmental and economic costs of AI expansion.
85 percent of enterprises run AI agents but few trust them
Eighty-five percent of enterprises are running AI agent pilots but only 5 percent have moved those agents into production according to Cisco's Jeetu Patel. The problem is the absence of a trust architecture for AI agents. Patel compared agents to teenagers who are supremely intelligent but have no fear of consequence. Cisco launched Defense Claw at RSAC 2026 to address the trust deficit and protect agents from the world and the world from agents.
AI compute crunch affects the entire economy
The AI compute crunch is affecting the labor market, gadget market, and electricity prices. AI companies are restricting access to their products, shutting down products, and beginning to increase prices. RAM, graphics cards, and storage for consumers have skyrocketed in price and are sold out in many stores. The age of cheap AI appears to be ending as venture capitalists and investment firms face higher costs to subsidize AI products.
Matt Pocock shares tips for effective LLM planning
AI expert Matt Pocock shared insights on effective LLM planning at an AI Engineer Europe talk. He introduced the concept of a smart zone and dumb zone for LLMs where the model performs best with clear concise prompts. Pocock advocated for a multi-phase plan approach that divides large tasks into smaller sequential steps. He also demonstrated his grill-me skill which helps developers stress-test an AI's understanding of a plan.
Apprenticeship.com launches job board as AI reshapes hiring
Apprenticeship.com announced its new job board for apprenticeships and paid work-based training opportunities. The platform connects job seekers with thousands of apprenticeship programs in skilled trades healthcare technology and business. The launch comes as artificial intelligence reshapes entry-level work and accelerates demand for skills-based hiring. The platform is free to use for career seekers.
Sources
- China's DeepSeek releases preview of long-awaited V4 model as AI race intensifies
- DeepSeek releases new flagship open source AI model V4
- China's DeepSeek rolls out a long-anticipated update of its AI model
- DeepSeek’s Sequel Set to Extend China’s Reach in Open-Source A.I.
- Breaking | DeepSeek releases next-gen AI model with ‘world-leading’ efficiency
- China's DeepSeek Rolls Out a Long-Anticipated Update of Its AI Model
- Neukgu: South Korea police arrest man over AI image of runaway wolf
- Man faces 5 years in prison for using AI to fake sighting of runaway wolf
- Tennessee Tech professor chosen as national AI education ambassador
- Therapy company mixes emotional and artificial intelligence to top ranking
- The man who is paying to see the future
- 'AI faces a growing backlash from the American public' due to its energy consumption and impact on jobs
- 85% of enterprises are running AI agents. Only 5% trust them enough to ship.
- The AI Compute Crunch Is Here (and It's Affecting the Entire Economy)
- Matt Pocock on LLM Planning: "Don't Bite Off More Than You Can Chew"
- Apprenticeship.com announces new Apprenticeship and Paid Training Job Board as AI Reshapes Entry-Level Hiring
Comments
Please log in to post a comment.