OpenAI released its o4-mini inference model via public API, featuring three "AI effort levels" (low/medium/high) for dynamic cost-performance optimization. The model processes complex scientific queries 40% faster than o3-mini with 220ms latency, now accessible to free-tier ChatGPT users with usage caps.
Google's Project Astra capabilities now power Gemini Live's screen-sharing function, enabling real-time AI assistance for coding and design tasks. The update achieves 95% accuracy in identifying on-screen objects and debugging code during developer beta tests.
Elon Musk's xAI reduced Grok API costs to $0.001 per 1K input tokens, making it the most affordable high-performance model. Early enterprise users report 30% savings on large-scale document analysis workflows.
Adobe opened Firefly Video beta access, enabling text-to-video generation with object-level motion control. Users can animate specific elements (e.g., "float the car") while freezing backgrounds, generating 15-second Hollywood-grade clips in under 2 minutes.
NVIDIA's Blackwell Ultra processors deliver 4x faster training for trillion-parameter models. Featuring quantum-AI hybrid cores, they accelerate molecular simulations by 20x for pharmaceutical breakthroughs.
Meta's latest Code Llama scored 92% on the SWE-bench coding benchmark, outperforming 85% of human developers. The open-source model supports 18 languages including Rust and COBOL, with enterprise deployment tools launching next week.
AlphaFold 4 now models 3D structures of RNA-protein interactions with 98% accuracy validated in peer review. The breakthrough accelerates drug discovery for neurological diseases like Parkinson's.
Stability Audio 2.0 enables voice cloning with 3-second samples and genre style transfer (e.g., "reggae Beethoven"). The tool watermarks outputs to prevent deepfake misuse while offering commercial usage rights.
200 Optimus Gen 2 robots began sorting packages at Amazon's Dallas facility, demonstrating L4 autonomy in logistics. Equipped with tactile sensors, they handle fragile items at human speed, cutting operational costs by 60%.
Amazon's AI agent audits cloud infrastructure in real-time, identifying 37% unused resources. Using reinforcement learning, it auto-optimizes configurations for Fortune 500 companies without human intervention.
IBM's toolkit automatically audits AI systems for bias and security risks, generating compliance reports. Integrated with Llama 4, it reduces legal review from weeks to hours for multinational corporations.
Hugging Face's new service offers free on-demand inference for 100+ models like Mistral 8x22B. APIs auto-scale during peak loads, democratizing billion-parameter AI access for researchers.
Siri now leverages Gemini for real-world visual analysis, answering queries like "What's wrong with this circuit?" with AR annotations overlaying physical components in real-time.
Palantir's AI simulated real-time battlefield strategies during NATO exercises, outperforming 92% of human commanders. It processes satellite/drone data to reduce collateral damage by 45%.
Perplexity's new tool generates citation-ready literature reviews from 100+ sources including arXiv. It traces sources and cuts research preparation from months to days for scientists.
See More Content about AI NEWS