Briefing

Sammanställd nyhet

ai-dev
OpenAI

Add GPT‑Realtime‑2 to your API to enable live voice reasoning, tool calling, and context up to 128K.

What to do now

Integrate GPT‑Realtime‑2 into your voice application to leverage live reasoning and tool calls.

Summary

Ingen sammanfattning tillgänglig

Key changes

  • GPT‑Realtime‑2 supports GPT‑5‑class reasoning, 128K context, parallel tool calls, preambles, adjustable reasoning levels
  • GPT‑Realtime‑Translate translates live from 70+ input languages to 13 output languages with lower WER in Hindi, Tamil, Telugu
  • GPT‑Realtime‑Whisper streams speech‑to‑text with low latency for live captions and notes
  • All models support real‑time audio input, live responses, and can be used in voice‑to‑action, systems‑to‑voice, voice‑to‑voice patterns
  • GPT‑Realtime‑2 includes stronger recovery, domain understanding, tone control, and adjustable reasoning effort
  • Safety features include active classifiers and custom guardrails via Agents SDK
  • GPT‑Realtime‑Translate offers 12.5% lower WER in key Indian languages
  • GPT‑Realtime‑Whisper can power live captions, meeting notes, and voice agents

Affects

internal

Source angles · 5 perspectives

OpenAI Blog
Vendor angle

Advancing voice intelligence with new models in the API

Open
Latent Space
Independent angle

OpenAI Launches GPT‑Realtime‑2 and Companion Voice Models

Open
AI News
Independent angle

GPT-Realtime-2, -Translate, and -Whisper: new SOTA realtime voice APIs

Open
Simon Willison
Independent angle

Quoting Luke Curley

Open
Hacker News (front page)
Independent angle

OpenAI’s WebRTC Design Flaws Make Voice AI Unreliable

Open

Customer impact

Analyzing matches…

Ask about this story

Impact on an agency? Which customers? Compare historically Risks of waiting