Build Hour: GPT-Realtime-2

The Build Hour format is doing real work here: it is not a research release or a product GA announcement, it is a structured demo designed to generate developer momentum before the API is widely stress-tested in production. The latency figure of sub-100ms is a marketing target, not a published benchmark under realistic network conditions.

Modelwire has no prior coverage to anchor this to directly, so context has to come from the broader space. OpenAI has been running Build Hours as a low-friction launch vehicle for several months, and GPT-Realtime-2 follows the same pattern as earlier realtime API previews: announce capability, show a demo, let developers find the edges. The competitive pressure here is real, with Google and ElevenLabs both shipping voice infrastructure, but the specific claims in this session have not been independently validated. Until developers report latency figures from actual API calls rather than a controlled demo environment, the sub-100ms headline should be treated as aspirational.

Watch whether independent developers post reproducible latency measurements from the GPT-Realtime-2 API within the next four to six weeks. If real-world p95 latency consistently clears 150ms rather than the stated sub-100ms, the enterprise workflow pitch loses its core technical premise.