GPT-Realtime-2.1
OpenAI's latest Realtime API model — better silence handling, noise robustness, and interruption behavior.
About
GPT-Realtime-2.1 is OpenAI's newer-generation Realtime API model, listed in OpenAI's model catalog by October 2026. It improves on the original gpt-realtime (now deprecated) in silence handling, noise robustness, and interruption behavior for turn-based voice agents.
It sits in the Realtime API family alongside GPT-Live-1: where GPT-Live-1 is the full-duplex voice layer with a delegated backend, GPT-Realtime-2.1 is the conversational realtime model evaluated head-to-head on the same speech benchmarks. OpenAI recommends marin and cedar among built-in voices for overall quality, and documents custom voice creation for eligible API customers.
In Artificial Analysis's Big Bench Audio run cited by xAI, GPT-Realtime-2.1 (High) averaged $0.055 cost per task, against $0.025 for Grok Voice 3.0 and $0.030 for GPT-Live-1 (Astra, Medium).
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent (Artificial Analysis Big Bench Audio, cited Oct 2026): $0.055 avg cost per task at High setting vs $0.025 Grok Voice 3.0, $0.030 GPT-Live-1 (Astra, Medium). Provider-reported: improved silence handling, noise robustness, and interruption behavior vs original gpt-realtime.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Existence and positioning (OpenAI Realtime API model; improves silence handling, noise robustness, interruption vs deprecated original gpt-realtime; listed in OpenAI's catalog by October 2026) corroborated by the OpenAI developer community summary of official announcements and independent voice-agent coverage. Docked for: (1) release date not pinned down in research, so releasedAt is left null; (2) capability claims come via community writeup of OpenAI docs rather than a dedicated launch announcement; (3) proprietary with no weights.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Adoption 6: OpenAI Realtime API model; supersedes deprecated gpt-realtime
- · Maintenance 9: listed in OpenAI catalog by Oct 2026
- · Openness 2: proprietary, no weights
- · Manual 7: existence and positioning corroborated by developer-community summary of official announcements and independent voice coverage; docked for unpinned release date (releasedAt null), no dedicated launch announcement, closed weights