China launched DeepSeek-R2😱

In the ever-accelerating race for AI dominance, the global spotlight is shifting—fast. From East to West, a new generation of large language models is reshaping the future of intelligent systems.

In China, DeepSeek-R2 emerges not just as a model, but as a national statement—an ambitious leap in multilingual reasoning and multimodal understanding, now powering everything from smart homes to robotics with unprecedented efficiency.

Meanwhile, across the enterprise world, Palmyra X5 is quietly rewriting the rules of what scalable AI can do. With a million-token context window and lightning-fast processing, this LLM from Writer is already redefining productivity for businesses on AWS Bedrock—at a fraction of the cost of its American rivals.

And Baidu? They’re not sitting still. The tech giant just unleashed ERNIE X1 Turbo and ERNIE 4.5 Turbo, delivering multimodal intelligence at breakthrough affordability—just a quarter the cost of DeepSeek R1.

Together, these updates signal a seismic shift in AI: where scale meets speed, where innovation crosses borders, and where tomorrow’s smartest machines are being built—today.

China’s Next-Gen AI Model Set to Rival Global Leaders

DeepSeek-R2 is a next-generation artificial intelligence model developed by the Chinese startup DeepSeek, positioned as a major competitor to leading Western AI technologies like GPT-4 and Claude, with a focus on advanced multilingual reasoning, enhanced code generation, and robust multimodal capabilities that allow it to process text, images, audio, and basic video. This model stands out for its efficient use of computational resources, innovative training techniques such as Generative Reward Modeling and Self-Principled Critique Tuning, and its ability to maintain high performance across multiple languages, especially Chinese and English, without the performance drop-off seen in many Western models. DeepSeek-R2 is already being integrated into consumer products from major Chinese brands like Haier and TCL, powering features in smart home appliances and robotics, and reflects China’s growing confidence and ambition in the global AI race by prioritizing open research, technological independence, and long-term goals like Artificial General Intelligence over immediate commercial gains.

Introducing Palmyra X5, the 1M-Token LLM for Scalable Enterprise Agents

Palmyra X5 is a powerful new enterprise large language model (LLM) launched by Writer, now available on AWS Bedrock, designed specifically for large-scale business applications with a massive one-million-token context window and ultra-fast, cost-efficient performance. It enables enterprise teams to build agents that can process huge volumes of domain-specific data, invoke multiple tools and workflows, and deliver near-instant responses, all while maintaining strict backward compatibility and predictable behavior. Palmyra X5 stands out with its hybrid attention mechanism for handling long inputs, Mixture of Experts (MoE) for smart resource allocation, and leading benchmark scores in reasoning, retrieval, and professional knowledge tasks. It processes a million-token prompt in just 22 seconds and handles multi-turn function calls in about 300 milliseconds, at a fraction of the cost of GPT-4.1, making it ideal for complex enterprise use cases like regulatory analysis, knowledge management, customer insights, and real-time orchestration. With seamless integration on both the Writer platform and AWS Bedrock, Palmyra X5 empowers businesses in regulated industries to deploy reliable, scalable, and auditable AI agents that can transform operations and decision-making.

Baidu Launches ERNIE X1 Turbo & 4.5 Turbo

Baidu has introduced ERNIE X1 Turbo and ERNIE 4.5 Turbo, the latest upgrades to its ERNIE AI model lineup, offering significantly faster performance and lower costs compared to previous versions. Both models are distinguished by their strong multimodal capabilities, robust reasoning skills, and cost-effectiveness, making them highly attractive for enterprise applications. Notably, ERNIE X1 Turbo is priced at just $0.14 per 1 million input tokens and $0.55 per 1 million output tokens, which is approximately 25% of the cost of DeepSeek R1, positioning it as one of the most affordable large language models on the market. These enhancements reflect Baidu’s commitment to advancing AI efficiency and accessibility, continuing the momentum established by the original ERNIE X1 and 4.5 models, and reinforcing Baidu’s leadership in the AI space.

Hand Picked Video

In this video, we'll look at how you can run DeepSeek-r1 on your Android Device.

Top AI Products from this week

• Breaktime Buddy - Break language barriers in video calls with AI-powered real-time voice translation. Ztalk is a desktop app that works with all video conferencing tools including Gmeet, Zoom and Teams.

• FullReviews - mrge is an AI-powered code review platform that automatically reviews PRs and gives human reviewers superpowers. It’s tool of choice for fast-moving teams like cal.com and n8n.

• TurboSQL - TurboSQL is a blazing-fast, AI-powered and beautiful SQL desktop app. With AI features so good, you might never have to write SQL again. Meet the Cursor for SQL databases.

• HelloCSV - A free FlatFile alternative - Linda is your personal interviewer & podcast host, guiding you through personalized interviews and transforming your conversations into short, shareable, professional-grade podcast episodes. No tech skills needed - just talk, and she'll take care of the rest.

• Helsa Health - Helsa is your personalized metabolic health coach—track diet, sleep, stress, activity, body, and blood markers. Get holistic, personlized, AI-driven science backed guidance to manage blood sugar, weight, and conditions like prediabetes, obesity etc.

• Transkriptor 2.0 - Transkriptor is your AI-powered note-taking and transcription tool — now smarter and faster with its latest version. Capture conversations, generate summaries, and extract insights effortlessly.

This week in AI

• Shop Faster - Find, compare, and buy products in ChatGPT with images, prices, reviews, and direct links. Independent results, no ads. Rolling out now to all users worldwide.

• AvatarFX Launc - Character.AI’s AvatarFX animates images into lifelike talking, singing videos with synced gestures. Safety steps include watermarking, face scrambling, and bans on minors’ images.

• Gmail iOS Redesign - Google rolls out Material 3 for Gmail on iPhone/iPad: pill-shaped search bar, blue accents, new Compose button, and modern look matching Android and web148

• AI Interpretability - Dario Amodei stresses urgent need to understand AI’s inner workings to reduce risks like deception, misuse, and errors. Progress in mechanistic interpretability offers hope to “steer the bus.

• AI Interpretability Urgency - Dario Amodei urges fast progress in AI interpretability to understand, control, and reduce risks like deception, misuse, and errors before AI grows too powerful.