Haberler
Google DeepMind Unveils New AI Model for Protein Structure Prediction with 90% Accuracy Boost
15.03.2026 21:35 | AIGoogle DeepMind has released a major upgrade to its AlphaFold system, now achieving over 90% accuracy in predicting protein structures from amino acid sequences — a 30% improvement over the previous version. The new model, AlphaFold 3, integrates diffusion-based generative modeling to predict interactions between proteins, DNA, RNA, and small molecules, significantly advancing drug discovery and molecular biology research. The updated model is now available to researchers via the AlphaFold Protein Structure Database.
OpenAI Announces GPT-4o Mini, a Lightweight, Cost-Effective Model for Edge Devices
15.03.2026 21:35 | AIOpenAI has launched GPT-4o Mini, a smaller, faster, and more affordable version of its flagship GPT-4o model, optimized for deployment on edge devices and low-resource environments. With 40% fewer parameters and 60% lower inference cost, GPT-4o Mini maintains strong performance on reasoning, coding, and multilingual tasks. It is now available via OpenAI’s API and will power future Apple and Samsung AI features starting next quarter.
EU Approves First AI Act Enforcement Guidelines for High-Risk Systems
15.03.2026 21:35 | AIThe European Union has finalized and published the first set of enforcement guidelines under its landmark AI Act, targeting high-risk AI systems in healthcare, critical infrastructure, and law enforcement. The guidelines, issued by the European Commission, mandate transparency logs, human oversight protocols, and bias audits for all deployed systems by October 2024. Non-compliant companies face fines up to 7% of global revenue.
NVIDIA Unveils Blackwell Ultra GPU with 200+ TFLOPS AI Performance for Next-Gen LLM Training
15.03.2026 21:35 | AINVIDIA has introduced the Blackwell Ultra GPU, the latest addition to its AI accelerator lineup, delivering over 200 TFLOPS of AI performance and 50% higher memory bandwidth than its predecessor. Designed for training trillion-parameter models, the Blackwell Ultra will power next-generation AI data centers and is expected to be integrated into Microsoft Azure and AWS cloud platforms by Q4 2024. NVIDIA also announced a new AI software stack, NeMo 2.0, optimized for the chip.
Anthropic Releases Claude 3.5 Sonnet with Enhanced Reasoning and Real-Time Web Browsing
15.03.2026 21:35 | AIAnthropic has launched Claude 3.5 Sonnet, a mid-tier model in its Claude 3 series that outperforms GPT-4o in coding benchmarks and exhibits significantly improved reasoning capabilities. The model features real-time web browsing, multi-modal understanding, and a 200K token context window. It is now available to enterprise users via Claude.ai and AWS Bedrock, with free tier access rolling out to all users within 48 hours.
Google DeepMind Unveils New AI Model for Scientific Discovery in Nature
06.03.2026 16:25 | AIGoogle DeepMind has released a new AI model called “AlphaFold 3,” capable of predicting the structure and interactions of proteins, DNA, RNA, and small molecules with unprecedented accuracy. Published in Nature, the model demonstrates a 50% improvement in predicting molecular complexes, accelerating drug discovery and biological research. The model’s code and weights are being made available to academic researchers.
OpenAI Launches GPT-4o Mini, a Low-Cost, High-Efficiency Model for Developers
06.03.2026 16:25 | AIOpenAI has introduced GPT-4o Mini, a smaller, faster, and more cost-efficient version of its flagship GPT-4o model. Designed for high-volume applications like chatbots and content moderation, GPT-4o Mini offers 90% of GPT-4o’s performance at 10% of the cost. It is now available via OpenAI’s API and integrates with ChatGPT for enterprise users.
EU Approves First AI Act Enforcement Guidelines for High-Risk Systems
06.03.2026 16:25 | AIThe European Union has finalized and published the first set of enforcement guidelines under its landmark AI Act, targeting high-risk AI systems in healthcare, education, and law enforcement. The guidelines, effective immediately, require transparency logs, human oversight, and bias audits for all deployed systems. Companies have six months to comply before penalties apply.
NVIDIA Unveils Blackwell Ultra GPUs for Next-Gen AI Data Centers
06.03.2026 16:25 | AINVIDIA has announced the Blackwell Ultra GPU, an enhanced version of its Blackwell architecture, delivering 25% more AI training throughput and 40% improved energy efficiency. The new chip is designed for large-scale generative AI workloads and will be available in Q3 2024. Major cloud providers, including AWS and Microsoft Azure, have already committed to deploying Blackwell Ultra in their data centers.
Anthropic Releases Claude 3.5 Sonnet with Enhanced Reasoning and Coding Capabilities
06.03.2026 16:25 | AIAnthropic has launched Claude 3.5 Sonnet, the latest model in its Claude 3 series, featuring significant improvements in logical reasoning, multi-step coding tasks, and real-time web search integration. Benchmarks show it outperforms GPT-4o in code generation and math problem-solving. The model is now available on Claude.ai and AWS Bedrock, with free tier access for individual users.