OpenAI launches GPT-Realtime-2 reasoning voice model

// 4h agoMODEL RELEASE

OpenAI launches GPT-Realtime-2 reasoning voice model

OpenAI released GPT-Realtime-2 in the API, a realtime voice model that can reason through spoken requests, call tools, and keep conversations flowing naturally. The launch adds 128K context, longer sessions, and stronger recovery behavior for production voice agents.

// ANALYSIS

Good fit for voice agents and assistant UX. The main change is tool use plus GPT-5-class reasoning in a voice-native model, with longer sessions and better recovery for real-time workflows. Cost control still matters because audio pricing is premium.

// TAGS

openaigpt-realtime-2realtime-apispeechtool-usespeech-to-speechmodel-release

DISCOVERED

4h ago

2026-05-07

PUBLISHED

4h ago

2026-05-07

RELEVANCE

10/ 10

AUTHOR

bridgemindai

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

MODEL21m ago

OpenAI GPT-Realtime-2 Powers Live Translator Demo

This retweet points to a live demo showing GPT-Realtime-2 as the core model for a real-time translator workflow. The timing lines up with OpenAI’s new voice-model release that also introduces GPT-Realtime-Translate and GPT-Realtime-Whisper, so the post reads more like an early builder showcase than a standalone product announcement.

UPDATE30m ago

Genspark Realtime Voice gains GPT-Realtime-2

Genspark says its Call for Me Agent now runs on OpenAI's GPT-Realtime-2, upgrading the voice stack behind its hands-free task runner. The move should improve longer, tool-heavy conversations like calls, scheduling, messaging, and follow-through.

OPEN SOURCE36m ago

Qwen3.6 mini GGUFs trim MTP grafting

This project strips Qwen3.6 GGUFs down to just the MTP tensors needed by buzz’s grafting script. The result is two tiny donor files, about 900MB for the 35A3B variant and 451MB for the 27B variant, instead of full 38GB and 29GB downloads.