1659 articles
Nvidia just rolled out its Nemotron-Cascade-8B reasoning model on Hugging Face, showing off impressive benchmark numbers powered by its Cascade RL training method.
Meta has launched SAM Audio, a unified AI model that lets users isolate and edit specific sounds within complex audio mixtures using text, visual, and time-based prompts.
NotebookLM's traffic surged from 30 million to over 110 million visits in 2025, driven by its integration into Gemini web that lets users attach notebooks as live conversation sources.
OpenAI's GPT-5.2 Pro achieved the highest score on the Mensa Norway benchmark, leading a comparative ranking of advanced AI models by reasoning performance.
OpenAI launched ChatGPT Images using its GPT Image 1.5 model, delivering 4x faster generation, improved instruction following, and precise editing capabilities for all users.
OpenAI's latest image generator GPT-Image-1.5 just grabbed the number one ranking on Image Arena's leaderboard, jumping nine positions ahead of its predecessor and outscoring competitors with a 1344 Elo rating.
Similarweb data reveals Grok captured the highest new-user rate in November 2025, while Perplexity dominated the six-month average and ChatGPT showed the lowest ratio among major AI platforms
Google's Gemini has gone head-to-head with ChatGPT in total new users, surpassing it in September and November, according to Similarweb.
Manus has rolled out version 1.6 featuring its Max agent, which shows stronger task completion rates, broader development capabilities, and measurable performance improvements across key benchmarks.
New research breaks down how memory works in AI agents, offering a practical framework with three key categories that could shape the next generation of intelligent systems.
NOETIX introduces Hobbs W1, a humanoid service robot already functioning as a guide and receptionist in museums, government offices, and corporate buildings across real-world environments.
OpenAI rolled out December 2025 updates to its Realtime API audio models, slashing error rates and hallucinations while delivering stronger performance in transcription, speech generation, and real-time instruction following.
xAI's Grok Code Fast 1 claimed the top spot on weekly token-usage rankings, outpacing Gemini, Claude, and OpenAI models as AI adoption accelerates across the industry.
El Salvador's teaming up with xAI to roll out Grok across its entire public school system—over 5,000 schools and more than a million students getting AI-powered personalized tutoring by 2027.
Anthropic's latest flagship model outperforms competitors on SWE-bench and delivers major efficiency gains with one-third the token costs.
ByteDance's Seedream 4.5 has jumped to second place in the Artificial Analysis Image Editing rankings, sitting just behind Google's Nano Banana Pro with an ELO score of 1,197 while leaving major competitors in the dust.
KwaiKAT's KAT-Coder-Pro V1 just hit 64 on the Artificial Analysis Intelligence Index—the highest score ever for a non-reasoning model—matching several reasoning-based systems while using way fewer output tokens.
Researchers introduced Bolmo, a fully open byte-level language model family that rivals traditional subword models across major benchmarks while requiring less than 1% of typical training resources.
The Sansa benchmark reveals major differences in content filtering across leading AI models, with GPT-5.2 showing the highest restriction levels among systems tested.
Anthropic is building an Autopatch feature for Claude Code that automatically scans repositories for security vulnerabilities and applies fixes without manual intervention.
We use cookies to improve your experience on our site and to show you relevant advertising. To find our more, read our privacy policy and cookie policy
Peter Smith
Saad Ullah
Eseandre Mordi
Victoria Bazir
Sergey Diakov
Alex Dudov
Usman Salis
