GPT-4o Realtime
The model that made voice agents feel instant: OpenAI's speech-to-speech GPT-4o over a single WebSocket, powering Advanced Voice Mode and a generation of live-translation demos.
Retired
GPT-4o Realtime appears to be discontinued or shut down. This page is kept as an archive record and is excluded from search engines.
About
GPT-4o Realtime (gpt-4o-realtime-preview, launched October 1, 2024 with the Realtime API) was OpenAI's first end-to-end speech-to-speech model: instead of chaining STT, an LLM, and TTS, it reasons natively over audio in one streaming WebSocket/WebRTC session — speech in, speech out, with interruptions, function calling, and custom voices handled by the same model. It powered ChatGPT's Advanced Voice Mode and became the reference implementation for low-latency voice agents and live translation tools, with a 32K-token context and per-token pricing ($5/$20 per 1M text tokens at launch, with higher audio-token rates). OpenAI has since moved the line to the GA gpt-realtime models (August 2025), and pricing trackers report the original preview retired in 2026 — so treat this entry as the historically important, now-legacy checkpoint of the realtime-voice era rather than something to build on today.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- No standardized public benchmarks; quality evidence is deployment-based (Advanced Voice Mode) and third-party comparisons\n- Latency widely reported in the sub-second range for turn-taking (community/third-party reports)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 5/10
Scored 5/10 (TIER metrics, metrics-v1). Evidence: official OpenAI Realtime API docs; powered Advanced Voice Mode; pricing documented by third-party trackers. Strengths: defined the low-latency speech-to-speech category, native audio reasoning, strong ecosystem. Docked for: (1) closed and proprietary — no weights, no reproducible evals; (2) preview line retired/superseded by the GA gpt-realtime models, so maintenance is over; (3) quality evidence is deployment anecdote, not benchmarks. License: proprietary.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 2.0/10
- Openness · license + docs
- 1.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Realtime API launched Oct 1, 2024; 32K context; powered ChatGPT Advanced Voice Mode
- · Launch pricing $5/$20 per 1M text tokens (higher audio-token rates)
- · Preview line superseded by GA gpt-realtime / gpt-realtime-2 (Aug 2025); pricing trackers report retirement 2026-05-07, unconfirmed by OpenAI docs
- · No public benchmarks — quality evidence is deployment-based only