All articles tagged with #ai model
OpenAI previews Astra, its next major AI model designed to tackle long-running, multi-step problems; internal work claims ten significant advances in mathematics and theoretical computer science, with proofs formalized as Lean certificates, and potential release as GPT-5.7 or GPT-6 under a tiered access policy.
SpaceXAI released Grok 4.5, its most capable model yet and its first since going public and acquiring Cursor, pitched as a coding/automation tool for business use. Trained with Cursor, Grok 4.5 is claimed to outperform some rivals on engineering and knowledge tasks while being faster and cheaper, priced at $2 per million input tokens and $6 per million output tokens. It’s available in Grok Build, Cursor on all plans, and SpaceXAI’s console (not yet in the EU). While it trails the largest OpenAI/Anthropic models in peak performance, SpaceXAI positions Grok as a cost-effective enterprise option and notes potential revenue from compute capacity leasing.
SpaceXAI and Cursor plan to launch their first jointly developed AI model, Grok 4.5, as soon as Wednesday after delaying the rollout to improve efficiency; the model is designed for rapid information processing and aims to rival Anthropic’s Opus 4.8.
Chrome on macOS can download a large local Gemini AI model (~4GB). To stop this, turn off On-device AI in Chrome via chrome://settings/system, and, if desired, delete the local model folder at ~/Library/Application Support/Google/Chrome/Default/OptGuideOnDeviceModel to reclaim space. The article also notes that disabling Apple Intelligence can free additional disk space and points to Google Support for more details.
Meta unveiled Muse Spark, its first AI model developed under Alexandr Wang, aiming to narrow the gap with OpenAI and Anthropic. The model will power queries in the Meta AI app and Meta.ai, with plans to expand to Facebook, Instagram, and WhatsApp, and accepts voice, text, and image inputs but outputs text. It includes modes like a fast query mode and a 'shopping mode' tied to user data; an open-source version is planned. While competitive on several tasks, it lags in areas like coding. All flavors are free but may be rate-limited, and Meta’s privacy policy indicates broad data use for AI features.
Shazam! star Zachary Levi says he’s gobsmacked by Seedance 2.0’s rapid AI progress, posting clips and noting how quickly generative AI can produce content; the piece highlights Hollywood IP concerns and ByteDance’s pledge to strengthen safeguards after studios’ legal pressure.
Apple researchers have developed UniGen-1.5, a unified AI model capable of understanding, generating, and editing images within a single system, enhancing previous models with new editing capabilities and improved instruction alignment, achieving state-of-the-art performance on various benchmarks, though it still faces challenges with text generation and identity consistency.
Google's new Gemini 3 Flash AI model is faster and more cost-effective than previous models, achieving high reasoning scores and outperforming Gemini 2.5 Pro in speed, with applications in search, customer support, and everyday tasks.
Google removed its AI model Gemma from a developer platform after US Senator Marsha Blackburn accused it of fabricating criminal allegations against her, highlighting concerns over bias and misinformation in AI tools. Google acknowledged hallucinations as a known issue and clarified that Gemma is intended for developers and researchers, not general consumer use.
Google's Gemini 2.5 Computer Use model, now in preview, can interact with web and mobile interfaces for tasks like browsing, clicking, and typing, demonstrating strong performance in web and Android environments, and is used for UI testing and workflow automation.
A researcher has developed a low-cost air quality estimation system using a microcontroller with a camera and a trained AI model to analyze sky images, offering a novel approach that complements traditional sensors, though with some limitations in accuracy.
Claude Sonnet 4.5 is a state-of-the-art AI coding model that significantly improves reasoning, math, and computer use capabilities, with extensive safety and alignment enhancements, new features like checkpoints and a SDK for building AI agents, and broad applications across software development, legal, financial, and creative tasks.
DeepSeek's R1 AI model, designed for reasoning tasks like math and coding, was trained cost-effectively at around $294,000 using reinforcement learning without copying from other LLMs, and has undergone peer review, setting a new standard for transparency and innovation in AI development.
A new AI model has been developed that can predict susceptibility to over 1,000 diseases, potentially revolutionizing personalized medicine and early diagnosis.
A new AI-based model suggests that multiple sclerosis (MS) should be viewed as a continuous disease spectrum rather than distinct subtypes, potentially improving disease management and treatment access by focusing on individual disease states rather than traditional categories.