The biggest story today wasn’t a new model card so much as the fight over what a model card is supposed to prove. OpenAI’s GPT-6 Astra reignited the public “has AGI arrived?” argument, after Astra posted 62.7% on ARC-AGI-3 and 99.9% when preserving reasoning state across requests. But as OpenAI, Nvidia, researchers and benchmarkers traded interpretations, the center of gravity shifted from single performance numbers to yardsticks, harnesses and governance—leaving “AGI” to function more like a messaging label than a testable threshold.
The same tension—capability claims versus verifiable systems—showed up across the rest of the market. Mistral’s €3 billion Series D, led by Samsung Electronics at a post-money valuation above €21 billion, doubles down on sovereign, open-weight infrastructure, while Berlin’s sci2sci raised €1.2 million to trace AI-generated biopharma claims to their sources, expanding its Integrity Cortex approach toward auditable evidence in regulated sectors. Veridue similarly targets workflow trust, pitching AI to compress renewable-energy and data-centre due diligence from 4-to-10 weeks into a few hours, using a deterministic agent layer to reduce the hand-wavy parts of deals.
Even OpenAI’s own leadership added friction: chief scientist Jakub Pachocki urged major labs to voluntarily slow research until safety standards catch up. If GPT-6 Astra made the dispute louder, today’s funding and tooling made it harder to ignore.
