GPT-Realtime-2 × BibiGPT

OpenAI launched GPT-Realtime-2, GPT-Realtime-Translate and GPT-Realtime-Whisper on 2026-05-07 — a voice-intelligence API trio with 128K context (up from 32K), GPT-5-class reasoning, real-time 70+→13 language translation and streaming Whisper STT. BibiGPT pairs the new endpoints for long-video subtitle generation, multilingual translation and Agent follow-up Q&A — without writing migration code yourself.

Generate subtitles with BibiGPT
Released · 2026-05-07 128K context · GPT-5-class Translate $0.034/min · Whisper $0.017/min

Key facts (90-second read)

As of 2026-05-09, OpenAI launched GPT-Realtime-2, GPT-Realtime-Translate and GPT-Realtime-Whisper on 2026-05-07 — a voice-intelligence API trio with 128K context (up from 32K), GPT-5-class reasoning, real-time 70+→13 language translation and streaming Whisper STT. Pricing: Realtime-2 at $32/$64 per MTok, Translate at $0.034/min, Whisper at $0.017/min. BibiGPT's routing layer is rotating the new endpoints into long-video subtitle generation, multilingual translation and Agent follow-up Q&A.

Features

What is GPT-Realtime-2?

OpenAI's 2026-05-07 voice-intelligence API release — three new endpoints (Realtime-2, Realtime-Translate, Realtime-Whisper) with 128K context, GPT-5-class reasoning and per-minute pricing for translation and STT.

128K context window

Realtime-2 jumps from the previous 32K cap to 128K tokens, enough to hold a full long-form lecture or a multi-hour podcast in a single voice session without chunking.

GPT-5-class reasoning over voice

OpenAI positions Realtime-2 as the voice-side counterpart of GPT-5 reasoning quality, with sharper multi-turn coherence and better tool calling than the previous Realtime model.

Real-time 70+→13 translation

Realtime-Translate accepts 70+ source languages, outputs 13 target languages and streams interpretation with low enough latency for live calls — priced at $0.034 per minute of audio.

Why this matters for BibiGPT users

BibiGPT routes long-form video subtitle generation, translation and Agent Q&A across multiple voice and ASR providers. A new Realtime API trio reshapes the routing for the hardest voice jobs.

Cheaper streaming subtitles

Realtime-Whisper drops streaming STT to $0.017 per minute — about half the cost of comparable real-time ASR. BibiGPT can lean on it for live YouTube / Bilibili / podcast subtitle pipelines.

One-step voice translation

Realtime-Translate folds STT + translation + TTS-style streaming into one endpoint. BibiGPT's translation pipeline can collapse the chain on supported language pairs for cleaner output.

Long-context voice Q&A

128K voice context lets BibiGPT's Agent answer follow-up questions over a 90-minute lecture in one session — without resummarizing or losing earlier-minute claims.

5 key changes (90-second read)

Headline shifts from the OpenAI voice-API release on 2026-05-07.

  1. 1

    Three new voice endpoints

    Realtime-2, Realtime-Translate and Realtime-Whisper ship as a trio. Callers pick the endpoint per use case instead of one general voice API for everything.

  2. 2

    Context jumps 32K → 128K

    Realtime-2 holds 4× more voice context. Long lectures, multi-hour podcasts and full meetings fit in one session without chunking or context-loss seams.

  3. 3

    GPT-5-class reasoning on voice

    Realtime-2 is positioned as the voice-side counterpart of GPT-5. Multi-turn voice agents, tool calling and structured retrieval get the same reasoning lift.

  4. 4

    Translate at $0.034/min, STT at $0.017/min

    Realtime-Translate covers 70+ source → 13 target languages and bills per audio minute. Realtime-Whisper streaming STT is roughly half the previous Realtime ASR price.

  5. 5

    Routing-layer absorbed for BibiGPT users

    If you use BibiGPT instead of integrating OpenAI directly, the routing layer rotates Realtime-2 / Translate / Whisper into video subtitles and translation. End users see better output without writing migration code.

3 typical scenarios for BibiGPT users

Where the new voice API trio pays off most for BibiGPT's user base.

Long-form video subtitle generation

A 90-minute Bilibili lecture or a 2-hour YouTube podcast. Realtime-Whisper streaming STT at $0.017/min cuts subtitle cost roughly in half versus the previous generation. BibiGPT routes the audio track through the new endpoint for cheaper, faster subtitles end to end.

Live multilingual translation

ja → en for technical talks, zh → ko for product reviews, en → zh-TW for legal explainers. Realtime-Translate folds STT + translation into one streaming endpoint at $0.034/min. BibiGPT's translation pipeline can use it on supported pairs for cleaner, lower-latency output.

Agent follow-up Q&A over a long video

Once BibiGPT has a summary, users ask voice-driven follow-ups: "what did the speaker say at minute 47 about pricing?". 128K voice context plus GPT-5-class reasoning lets the Agent answer over the full lecture in one session — no resummarization, no lost earlier-minute claims.

Frequently Asked Questions

Ask us anything!

Use BibiGPT for video subtitles & translation — backed by Realtime-2-tier voice models

BibiGPT auto-routes between OpenAI Realtime, Anthropic and Gemini for video subtitle generation, multilingual translation and follow-up Q&A. You get the right voice model for the job without managing migrations or per-minute billing yourself.

Try BibiGPT free