Chief executive Sam Altman told staff the model will reach a limited enterprise preview first, with federal officials approving customer access on a case-by-case basis.
Anthropic disabled both frontier models globally on June 12 to comply with a Commerce Department order barring foreign-national access, leaving subscribers without the systems that shipped nine days earlier.
Less than two years after Google paid roughly $2.7 billion to rehire him from Character.AI, Shazeer will lead architecture research for next-generation models.
An independent legal research study shows new reasoning architectures match human accuracy in identifying prior-art conflicts.
Federal agencies may now deploy the model inside IL5 enclaves for document review and analyst workflows, a milestone rival labs have not matched at equivalent clearance levels.
The company's new model scores 94.7% on the GPQA Diamond benchmark — a suite designed to defeat AI systems — marking the first time a model has exceeded the median performance of PhD-level domain experts.
Meta's latest open-weight release achieves state-of-the-art performance on coding, mathematical reasoning, multilingual understanding, and scientific problem-solving, while being available for commercial deployment under a modified community license.
Free Morning Briefing
A calm, intelligent briefing delivered daily before the market opens.
Free. No spam. Unsubscribe anytime.
