DeepSeek has released two new AI models, the V4 Pro and V4 Flash, at notably low prices. The V4 Pro boasts 1.6 trillion total parameters with 49 billion active, while the V4 Flash has 284 billion total and 13 billion active. Both models support a 1 million token context window and are open source. They perform close to top models like GPT 5.4 and Claude Opus 4.6, with the V4 Pro leading in coding benchmarks and agentic tasks, though it trails on factual knowledge retrieval. DeepSeek estimates it is 3 to 6 months behind top US AI labs.
NVIDIA announced that its Blackwell platform supports the DeepSeek V4 models for fast AI inference. On the NVIDIA GB200 N72, the V4 Pro achieves over 150 tokens per second per user. The V4 family uses hybrid attention to reduce memory and computation costs for long context inference. Meanwhile, Google split its AI chips into separate training and inference TPUs at Google Cloud Next 26, announcing the TPU v8t for training and the TPU v8i for inference to boost performance and energy efficiency.
Anthropic added integrations for Claude with apps including Spotify, Uber, Uber Eats, Instacart, and Booking.com. Claude now suggests the right app based on the user's conversation and checks before completing any booking or purchase. Claude has over 200 integrations and competes with OpenAI and Google. CrowdStrike launched Project QuiltWorks, a coalition with Accenture, EY, IBM Cybersecurity Services, Kroll, and OpenAI, to help organizations find and fix vulnerabilities in production code that advanced AI models can uncover.
Alibaba announced that its Qwen AI model will be integrated into vehicles from multiple Chinese automakers, including Audi and BMW. The system runs on Alibaba Cloud and allows drivers to use voice commands to order food, book hotels, make payments, and navigate. The announcement was made at the Beijing Auto Show 2026 as Chinese automakers add AI features to attract buyers in a slowing electric car market.
Old industrial sites in Britain, like Wilton International in Teesside, are becoming hotspots for AI data centers due to their existing power plants and grid connections. Demand for grid connections has surged 460% in early 2025, with wait times stretching to 12 to 15 years. AI data centers can be built far from London because they mainly need processing power, not speed.
Project Liberty Institute held events in Seoul and Tokyo to promote responsible AI investment. A white paper found that nine out of ten venture capitalists see financial opportunity in responsible AI. The institute aims to bring Asian investors into global conversations about AI norms. Penn State launched an online AI Essentials course for its employees, covering technical knowledge, ethics, critical thinking, and practical applications, alongside a new AI platform called AI Studio.
A debate over AI in schools highlights benefits and risks. A student argues AI enables personalized learning and quick feedback, while teachers warn it can erode basic math skills and encourage cheating. Some educators compare using AI to complete assignments with academic dishonesty, and the debate centers on whether AI should be banned or taught responsibly.
Key Takeaways
- DeepSeek released V4 Pro (1.6 trillion total parameters, 49 billion active) and V4 Flash (284 billion total, 13 billion active) at low prices, both open source with 1 million token context windows.
- DeepSeek V4 Pro leads in coding benchmarks and agentic tasks but trails on factual knowledge retrieval and has a 94% hallucination rate when it doesn't know an answer.
- DeepSeek estimates it is 3 to 6 months behind top US AI labs.
- NVIDIA's Blackwell platform supports DeepSeek V4 models, achieving over 150 tokens per second per user on the GB200 N72.
- Google split its AI chips into separate TPU v8t for training and TPU v8i for inference to boost performance and energy efficiency.
- Anthropic added integrations for Claude with Spotify, Uber, Uber Eats, Instacart, and Booking.com, with over 200 total integrations.
- CrowdStrike launched Project QuiltWorks with Accenture, EY, IBM, Kroll, and OpenAI to fix vulnerabilities in production code.
- Alibaba's Qwen AI will be integrated into vehicles from Audi, BMW, and other Chinese automakers for voice-controlled tasks.
- Demand for grid connections for AI data centers in Britain surged 460% in early 2025, with wait times of 12 to 15 years.
- Project Liberty Institute found that nine out of ten venture capitalists see financial opportunity in responsible AI.
DeepSeek launches V4 Pro and V4 Flash AI models at low prices
DeepSeek released two new AI models, DeepSeek V4 Pro and DeepSeek V4 Flash. V4 Pro has 1.6 trillion total parameters with 49 billion active, while V4 Flash has 284 billion total with 13 billion active. Both models support a 1 million token context window and are open source. The models perform at levels close to top models like GPT 5.4 and Claude Opus 4.6 but cost much less. DeepSeek V4 Pro leads in coding benchmarks and agentic tasks, though it trails on factual knowledge retrieval.
DeepSeek says it is 3 to 6 months behind top US AI labs
DeepSeek had revealed two very capable V4 models today, and it’s even come up with an estimate of how far it is from the state-of-the-art frontier models. In its technical report accompanying the V4 release, DeepSeek states that V4-Pro-Max “demonstrates superior performance relative to GPT-5.2 and Gemini-3.0-Pro on sta...
DeepSeek V4 Pro ranks second among open models on AI index
DeepSeek V4 Pro scored 52 on the Artificial Analysis Intelligence Index, making it the second highest rated open weights reasoning model behind Kimi K2.6. V4 Flash scored 47. V4 Pro leads open weights models on agentic real world tasks with a score of 1554 on GDPval AA. However, both models have high hallucination rates, with V4 Pro hallucinating 94% of the time when it does not know an answer. The models are cheaper than top closed source models but more expensive than some open weights peers due to high token usage.
Old industrial sites in Britain become hotspots for AI data centers
Land left empty by the decline of the chemical industry in northeastern England is now valuable for AI data centers. Sites like Wilton International in Teesside have power plants and grid connections that tech giants need. Across Britain, owners of industrial sites, old factories, and even farms are trying to attract data center investment. The demand for grid connections has surged 460% in early 2025, with wait times stretching to 12 to 15 years. AI data centers can be built far from London because they mainly need processing power, not speed.
Google splits its AI chips into separate training and inference TPUs
At Google Cloud Next 26, Google announced two new eighth generation Tensor Processing Units: the TPU v8t for training and the TPU v8i for inference. This split is designed to boost performance and energy efficiency by optimizing each chip for its specific workload. The move signals a shift toward workload specialized AI infrastructure.
Penn State launches online AI Essentials course for employees
Penn State launched a new online course called AI Essentials for its employees. The course has four modules covering technical knowledge, ethics, critical thinking, and practical applications. It launched at the same time as AI Studio, a new AI platform for the university. The course is not required but is strongly encouraged, and it will be foundational for future AI literacy offerings.
NVIDIA Blackwell supports DeepSeek V4 models for fast AI inference
NVIDIA announced that its Blackwell platform supports DeepSeek V4 Pro and V4 Flash models. DeepSeek V4 Pro has 1.6 trillion total parameters and 49 billion active, while V4 Flash has 284 billion total and 13 billion active. Both models support up to a 1 million token context window. On NVIDIA GB200 NVL72, DeepSeek V4 Pro achieves over 150 tokens per second per user. The V4 family uses hybrid attention to reduce memory and computation costs for long context inference.
Alibaba brings Qwen AI to cars for voice controlled bookings and orders
Alibaba announced that its Qwen AI model will be integrated into vehicles from multiple Chinese automakers. The system runs on Alibaba Cloud and allows drivers to use voice commands to order food, book hotels, make payments, and navigate. Auto companies including Audi, BMW, and others will use the technology. The announcement was made at the Beijing Auto Show 2026. Chinese automakers are adding AI features to attract buyers in a slowing electric car market.
Debate over AI in schools highlights benefits and risks for students
A high school student and educators debate the role of AI in classrooms. The student argues AI enables personalized learning and quick feedback, helping students learn faster. Teachers warn that AI can erode fundamental skills like basic math and encourage cheating. Some educators compare using AI to complete assignments with academic dishonesty. The debate centers on whether AI should be banned entirely or taught responsibly to prepare students for an AI driven world.
Project Liberty Institute pushes for responsible AI investment in Asia
Project Liberty Institute held events in Seoul and Tokyo to promote responsible AI investment. In Seoul, a roundtable with Korean investors and policymakers discussed AI risks like governance failures and market concentration. In Tokyo, sessions focused on data related human rights risks and investor due diligence. A white paper found that nine out of ten venture capitalists see financial opportunity in responsible AI. The institute aims to bring Asian investors into global conversations about AI norms.
Anthropic connects Claude to popular apps like Spotify and Uber
Anthropic added integrations for Claude with apps including Spotify, Uber, Uber Eats, Instacart, and Booking.com. Claude now suggests the right app based on what the user is doing in a conversation. The system checks with the user before completing any booking or purchase. Claude has over 200 integrations and competes with OpenAI and Google. The connector model turns Claude into an orchestration layer that can handle tasks like ordering groceries or booking a ride without switching apps.
CrowdStrike launches Project QuiltWorks AI security coalition
CrowdStrike launched Project QuiltWorks, a coalition with Accenture, EY, IBM Cybersecurity Services, Kroll, and OpenAI. The initiative helps organizations find and fix vulnerabilities in production code that advanced AI models can uncover. It uses models from OpenAI and Anthropic along with CrowdStrike's own threat intelligence. The coalition assesses security posture, scans code, ranks findings by exploitability, and guides fixes. CrowdStrike also introduced a Frontier AI Readiness and Resilience Service for ongoing customer engagements.
Sources
- DeepSeek Releases DeepSeek V4-Pro & V4-Flash, Delivers GPT 5.4 & Opus 4.6-Level Performance At Fraction Of The Price
- DeepSeek Says It Is 3-6 Months Behind US Frontier Labs
- DeepSeek V4 Pro Becomes Second-Highest Rated Open Model On Artificial Analysis Index With Score Of 52
- Powered Land and Zombie Projects: Real Estate in the Age of AI
- Google splits AI chips into training and inference TPUs, signaling shift toward workload-specialized AI infrastructure
- Penn State launches new online ‘AI Essentials’ literacy course for employees
- Build with DeepSeek V4 Using NVIDIA Blackwell and GPU-Accelerated Endpoints
- Alibaba’s Qwen AI is coming to cars, allowing drivers order food and book hotels by voice
- AI has drawbacks, but we shouldn't abolish it from schools entirely
- Advancing Responsible AI Investment Across Asia
- Anthropic Turns Claude Into a Front Door for Daily Apps
- CrowdStrike launches AI security coalition with partners
Comments
Please log in to post a comment.