Categories
AI News

x = y Is Where Even Frontier Models Lose Track, Says Ai2’s StateBench

Ai2’s open StateBench tests how models track changing values in long text. Frontier models fall to 13% or lower on the hardest level.

Categories
AI News

FlashProve Can Prove Which Model Answered You. Slowly, for 1,024 Tokens

Anthropic’s open FlashProve proves which model wrote a reply without revealing its weights. It costs ~70x the compute and stops at 1,024 tokens.

Categories
AI News

An Open Model Learned Science From Lab Notebooks, Not Just Papers

Deep Principle’s open-weight Hydrogen-1.0 learned from lab notebooks and says it beats GPT-6-Astra at research science. The scores are its own.

Categories
AI News iRaluca

Anthropic Will Ban Cruelty to Claude. Rudeness Is Still Allowed

From 12 November, Anthropic’s policy bans needless, sustained cruelty toward Claude. Frustration is fine. What the rule says, and why it’s contested.

Categories
AI News iRaluca

A 7.72 TB “Anthropic Leak” on Hugging Face Was Noise With a Name Tag

A 7.72 TB Hugging Face repo claimed to be a leaked Anthropic model. Volunteers rebuilt its “weights” from a random seed, bit for bit.

Categories
AI News iRaluca

GDPx Gives the Best AI Agent 12%. How the Rest Fail Says More

SemiAnalysis’s new GDPx benchmark: 206 agent tasks in 11 domains. Claude Opus 5.5 leads at just 12%. How the models fail is the real story.

Categories
AI News iRaluca

Pika Built Its Open Video Model on MiniMax’s, and Kept the 4K for Itself

Pika’s open H3 Flex adapters turn MiniMax’s open video model into a six-step clip maker. No benchmarks, and 4K stays behind Pika’s API.

Categories
AI News iRaluca

OpenAI’s New Biology Model Has Old Weights and a New Front Door

OpenAI’s GPT-Rosalind-6 uses GPT-6 Astra’s exact weights but refuses fewer biology questions, for vetted users only, at a higher price.

Categories
AI News

IBM Patched Granite Switch Like Software. The Base Model Never Moved

IBM’s open Granite Switch 4.1.1 swaps two plug-in parts and leaves the 30B base model untouched. A modular idea, still light on numbers.

Categories
AI News iRaluca

Four AI Labs Agree to Read One “Don’t Train on Me” Line in robots.txt

Google, Microsoft, OpenAI and Anthropic will read one robots.txt line asking them not to train on a site. It only looks forward.