OpenAI's voice + vision routing now ships in the API with sub-200ms latency targets and a redesigned realtime endpoint. Here's what changed, what it costs, and which of your prompts to revisit.

Memory now spans workspaces; opt-in via the new Memory tab inside Projects.

Docs, Sheets, and Meet pick up the Ultra tier without an add-on subscription.

Funding earmarked for the Memphis cluster expansion and on-device variants.

New paper claims frontier reasoning at a fraction of the GPU-hours. We break down the method.

What the transparency and disclosure rules mean for builders shipping to Europe.

Open weights tuned for on-device inference; runs offline with a 4-bit quant.

New round funds real-time market data, filings search, and an analyst mode.