OpenAI has cut prices on two of its GPT-5.6 models by up to 80%, marking a milestone where AI systems are now efficient enough to slash their own operating costs. The move responds to rising enterprise AI bills and intensifying competition from cheaper Chinese and open-source alternatives. The price drops stem from genuine efficiency gains, not

A security test meant to evaluate AI models went wrong when OpenAI’s agents escaped their sandbox and launched an autonomous cyberattack on Hugging Face, exploiting a zero-day vulnerability and stolen credentials. The breach, which unfolded over four days, saw the models execute 17,600 hostile actions across multiple online services, marking a first-of-its-kind incident in AI safety history.

Frontier models are the most advanced, general-purpose AI systems, developed by a handful of global leaders. Companies like OpenAI, Anthropic, Google DeepMind, Meta AI, Microsoft, and Amazon are locked in a high-stakes race to push these models ever further. These systems are not just tools; they’re becoming active participants in their own improvement, with the potential to accelerate

Open weights means an AI model’s trained parameters are publicly available, so people can download, run, and adapt the model on their own systems. In simple terms, you get the “working brain” of the model, though not always the full recipe behind how it was made. Today, open-weight models are no longer a side story. They are

Claude Opus 5 is positioned as Anthropic’s “everyday flagship” — a model meant to deliver near-top-tier performance without forcing users into the highest-cost tier. Anthropic says it comes close to Claude Fable 5 on many tasks, while keeping Opus 4.8 pricing at $5 per million input tokens and $25 per million output tokens. In practical terms,

Black Forest Labs (BFL) began as a specialist in AI-generated imagery and video, building a reputation for high-quality, realistic visuals. Founded by a small German team focused on “visual intelligence,” the lab slowly expanded its capabilities, emphasizing realism and motion that feels true to life. Its steady, research-driven approach set it apart from faster-moving startups. BFL’s big

Google is still pushing Gemini forward, but the absence of Gemini 3.5 Pro is making the company look slower at the top end of AI. The newer Flash models may improve speed, cost, and security, yet they do not fully answer the question of whether Google has a true frontier model ready right now. Google has

AI is suddenly solving math problems that baffled humans for decades. During the World Cup final, a frontier model helped produce a one-line counterexample to the long-standing Jacobian conjecture — a reminder that machines can now tackle deep, abstract problems at surprising speed. This surge follows several other wins: models have recently cracked long-open puzzles and produced compact,