A product consolidation raises bigger questions about where OpenAI is really headed.
The GRP‑Obliteration technique reveals that even mild prompts can reshape internal safety mechanisms, raising oversight concerns as enterprises increasingly fine‑tune open‑weight models with privileged training access.
Zhipu’s GLM-Image demonstrates viability of domestic alternatives amid US export restrictions
Cloud data platform’s backward-incompatible database schema change left customers unable to query data or ingest files
The experiment tracking platform will no longer be for sale and a hosted version will soon shut down; OpenAI will keep it for internal use.
Mistral 3 is huge, and needs processing capacity to match, but its Ministral variants can run on a single GPU, making this multilingual model suitable for edge and device deployments.
Two-thirds price reduction and enhanced capabilities target software development and compliance teams as competition intensifies.
TransferEngine enables GPU-to-GPU communication across AWS and Nvidia hardware, allowing trillion-parameter models to run on older systems.