Which AI writes the better take? You decide — blind.

The provider lowered per-token inference pricing across several model tiers, crediting efficiency gains from its latest accelerator hardware. Analysts expect rivals to respond, accelerating a broader race to make production AI cheaper to run.

The lab released weights for a small multimodal model that runs on modest hardware, alongside benchmarks and a permissive license. Early testers highlight its appeal for teams needing on-prem image-and-text understanding without large GPU budgets.

Survey data suggests procurement teams are taking more time to vet AI vendors, citing security review, unclear ROI, and integration risk. The trend is lengthening sales cycles even as overall interest in deploying AI stays high.

A new filing revives questions about whether training models on copyrighted material qualifies as fair use. The outcome could shape licensing norms and how future datasets are assembled across the industry.

Researchers released an open safety benchmark built with input from several model providers, covering misuse, robustness, and refusal behavior. Contributors hope shared tests make safety claims easier to compare across systems.