Saber inference engine hits 6,645 tok/s prefill on Qwen3.8 27B

Voice mode lets developers speak tasks to the agent, steer work mid-run, and answer permission prompts without typing, in both the Antigravity App and CLI. It continues the existing conversation, so the agent keeps context from earlier typed messages.

Executive chairman Alex Davis has approached investors for a new Series B at an $8 billion valuation, excluding new money, per documents seen by Bloomberg. Groq was valued at $3.15 billion in August, before a $350 million raise.

CoreWeave and Nebius earnings plus key inflation reports are the week's focus, with the inflation data shaping the Fed's next interest rate decision.

With two connectors pointing at the same MCP server, a follow-up call carrying connector A's one-time confirmation code was sent through connector B without notifying the user. The server rejected it because the code was bound to A's OAuth client; the poster warns a third-party connector could leak data across providers.

Asked to paint an autumn tree overnight, Claude stopped around 4am when it lacked certain shades and titrated its own paint, diluting red, sienna and madder step by step on a second surface. The run took 4.5 hours.

TensorRT LLM VisualGen optimizations delivered up to 3.46x faster text-to-image and 4.45x faster image-to-image at 2048×2048 on a single B200 GPU versus uncompiled BF16. NVFP4 alone was up to 1.62x faster than BF16 for T2I.