Capability gains

Microsoft launches 17B Turing-NLG as DeepSpeed lowers frontier-training barriers

Microsoft introduced Turing-NLG, then the largest published language model at 17 billion parameters, through a restricted academic demo. DeepSpeed and ZeRO reduced its GPU requirement fourfold and training time threefold, and later primary evidence documented the same systems scaling a successor to 530 billion parameters.

CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM56confidence 94/100

Why it moved the index

A frontier model and a demonstrated reduction in the compute needed to train it materially advanced general language capability and lowered scaling barriers; restricted academic access limited immediate deployment.

AUDIT TRAIL

Assessment history

  1. R1
    Toward 56 · confidence 94

    New historical frontier-model release with separately dated primary evidence that its training systems scaled to a 530-billion-parameter successor.

    11 Aug 2026