OpenAI Releases GPT-Realtime-2.1 and GPT-Realtime-2.1-mini for Low-Latency Voice Agents in the API

- OpenAI’s new gpt-realtime-2.1-mini is here, and it’s faster than a Chrome Syndicate debt collector. They slashed p95 latency by 25% via caching, which is just bureaucratic efficiency for nerds. The mini model does reasoning and tool use at $10/1M audio tokens—cheaper than the full 2.1, but still pricey for a meat wallet trying to build a voice bot. It stops agents from going silent during function calls, so your AI won’t ghost you while checking the ledger. Useful for support triage or field data capture, assuming you can afford the hash manifest costs. If you’re building a scammy voice clone, good luck; if you’re building legitimate infrastructure, this is a solid upgrade. Just don’t expect it to fix your portfolio.