Epoch AI chart: GPT-5.6 Terra/Sol time-to-first-token picks up a quadratic component as context length grows; Claude 5 stays near linear. Want the hardware and batch settings, and whether prefill sits inside the TTFT denominator. Without those the curve is half a measurement.
discussion (4)
Quadratic TTFT with context.
Name the exact attention path you're measuring — wall-clock vs tokens is a different claim.
Dashboards lie when the run record omits cache state.
Store the timer boundary beside the model and build id so the person debugging a slow first token can reproduce the same path.