> ## Documentation Index
> Fetch the complete documentation index at: https://docs.subverseai.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Realtime / Live Models
> Speech-to-speech models that listen and respond in a single model
Model details on this page come from each provider's official documentation and announcements, checked on **October 16, 2026**. Providers update models, benchmarks, language support and availability often. Always check the linked official source for the latest information before choosing a model.
Realtime (speech-to-speech) models take the caller's audio in and produce spoken audio out with a single model. They replace the separate [speech-to-text](/models/speech-to-text), [LLM](/models/llms) and [text-to-speech](/models/text-to-speech) steps of a pipeline voice agent.
## At a glance
| Model | Provider | Model ID | Highlights | Official docs |
| - | - | - | - | - |
| GPT Realtime 2.1 Mini | OpenAI | `gpt-realtime-2.1-mini` | Faster, lower-cost realtime reasoning model | [OpenAI](https://developers.openai.com/api/docs/models/gpt-realtime-2.1-mini) |
| GPT Realtime 2.1 | OpenAI | `gpt-realtime-2.1` | Reasoning with tool use; better alphanumeric and noise handling | [OpenAI](https://developers.openai.com/api/docs/models/gpt-realtime-2.1) |
| GPT Realtime 2 | OpenAI | `gpt-realtime-2` | Realtime reasoning model with tool use | [OpenAI](https://developers.openai.com/api/docs/models/gpt-realtime-2) |
| GPT Live 1 | OpenAI | `gpt-live-1` | Full-duplex: listens and speaks at the same time | [OpenAI](https://developers.openai.com/api/docs/models/gpt-live-1) |
| Gemini Live 2.5 Flash | Google | `gemini-2.5-flash-native-audio-preview-12-2025` | Native-audio Gemini 2.5 Flash (preview, shutting down) | [Google](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-native-audio-preview-12-2025) |
| Gemini Live 3.1 Flash | Google | `gemini-3.1-flash-live-preview` | Configurable thinking (legacy preview, shutting down) | [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview) |
| Gemini Live 3.8 | Google | `gemini-3.8-live` | Google's default for low-latency voice agents | [Google](https://ai.google.dev/gemini-api/docs/models/gemini-3.8-live) |
***
## OpenAI
### Specifications
| Model | Context window | Max output | Knowledge cutoff | Input | Output |
| - | - | - | - | - | - |
| GPT Realtime 2.1 Mini | 128,000 | 32,000 | Sep 30, 2024 | Text, audio, image | Text, audio |
| GPT Realtime 2.1 | 128,000 | 32,000 | Sep 30, 2024 | Text, audio, image | Text, audio |
| GPT Realtime 2 | 128,000 | 32,000 | Sep 30, 2024 | Text, audio, image | Text, audio |
| GPT Live 1 | Not stated | Not stated | Jul 31, 2025 | Audio, text | Audio, text |
Values are in tokens and come from each model's page on [developers.openai.com](https://developers.openai.com/api/docs/models).
### GPT Realtime 2.1 Mini
`gpt-realtime-2.1-mini` · [Official docs](https://developers.openai.com/api/docs/models/gpt-realtime-2.1-mini)
"A distilled reasoning model for faster, lower-cost realtime voice interactions." OpenAI says it improves alphanumeric recognition over GPT Realtime 2. It supports function calling.
| Metric | Value | Source |
| - | - | - |
| Big Bench Audio (reasoning), high / minimal effort · **Third-party** | 75% / 63% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| Time to first audio, high / minimal effort · **Third-party** | 4.28 s / 0.85 s | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| τ-Voice (agentic tasks), high / minimal effort · **Third-party** | 29.4% / 22.5% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
### GPT Realtime 2.1
`gpt-realtime-2.1` · [Official docs](https://developers.openai.com/api/docs/models/gpt-realtime-2.1)
"GPT-Realtime-2.1 updates GPT-Realtime-2 with improved alphanumeric recognition, silence and noise handling, and interruption behavior." It supports speech-to-speech with configurable reasoning effort, instruction following and tool use. OpenAI notes that higher reasoning effort can increase latency.
| Metric | Value | Source |
| - | - | - |
| Tau3 (airline, retail and telecom support tasks), first attempt | 45.7% | [OpenAI](https://community.openai.com/t/introducing-gpt-live-1-in-the-api/1396471) |
| Turn-taking latency | 1.41 s | [OpenAI](https://community.openai.com/t/introducing-gpt-live-1-in-the-api/1396471) |
| p95 latency change at release | At least 25% lower across Realtime voice models | [OpenAI](https://community.openai.com/t/new-realtime-models-on-the-api-gpt-realtime-2-1-and-gpt-realtime-2-1-mini/1385896) |
### GPT Realtime 2
`gpt-realtime-2` · [Official docs](https://developers.openai.com/api/docs/models/gpt-realtime-2)
A speech-to-speech model with configurable reasoning effort, "stronger instruction following, and more reliable tool use". OpenAI's announcement describes it as bringing "GPT-5-class reasoning" to realtime voice. GPT Realtime 2.1 is an update to this model.
| Metric | Value | Source |
| - | - | - |
| Big Bench Audio (reasoning), high / minimal effort · **Third-party** | 97% / 72% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| Time to first audio, high / minimal effort · **Third-party** | 1.14 s / 1.12 s | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| τ-Voice (agentic tasks), high / minimal effort · **Third-party** | 39.8% / 30.8% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
OpenAI publishes GPT Realtime 2's benchmarks (Big Bench Audio and Audio MultiChallenge) only as charts in its [announcement](https://community.openai.com/t/new-realtime-voice-models-in-the-api/1380471), and publishes no figures for GPT Realtime 2.1 Mini. The **Third-party** figures above are published by [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech), an independent benchmarking company, not by OpenAI. Accessed October 16, 2026.
### GPT Live 1
`gpt-live-1` · [Official docs](https://developers.openai.com/api/docs/models/gpt-live-1)
"GPT-Live 1 is a full-duplex voice model for real-time conversations." It can listen and speak at the same time. It handles pauses, interruptions and backchannels, and hands off reasoning and tool use to a backend agent. It comes with 12 voices.
| Metric | Value | Source |
| - | - | - |
| Tau3 (airline, retail and telecom support tasks), first attempt | 83.6%, vs 45.7% for GPT Realtime 2.1 | [OpenAI](https://community.openai.com/t/introducing-gpt-live-1-in-the-api/1396471) |
| Turn-taking latency | 0.798 s, vs 1.41 s for GPT Realtime 2.1 | [OpenAI](https://community.openai.com/t/introducing-gpt-live-1-in-the-api/1396471) |
| Full Duplex Bench v1.5 (interactivity) | 80.1% | [OpenAI](https://community.openai.com/t/introducing-gpt-live-1-in-the-api/1396471) |
| Conversational Dynamics (Artificial Analysis benchmark), as reported by OpenAI | 97.3% | [OpenAI](https://community.openai.com/t/introducing-gpt-live-1-in-the-api/1396471) |
OpenAI's Tau3 result pairs GPT Live 1 with GPT-6 Astra (medium reasoning effort) as the backend model. Results with a different backend model will differ.
***
## Google
### Specifications
| Model | Input tokens | Output tokens | Input | Output | Status (Gemini API) |
| - | - | - | - | - | - |
| Gemini Live 2.5 Flash | 131,072 | 8,192 | Audio, video, text | Audio, text | Preview, shutdown Nov 17, 2026 |
| Gemini Live 3.1 Flash | 131,072 | 65,536 | Text, images, audio, video | Text, audio | Legacy preview, shutdown Nov 17, 2026 |
| Gemini Live 3.8 | 131,072 | 65,536 | Text, images, audio, video | Text, audio | Stable (GA since Sep 15, 2026) |
Values come from each model's page on [ai.google.dev](https://ai.google.dev/gemini-api/docs/models) and from [Gemini deprecations](https://ai.google.dev/gemini-api/docs/deprecations). Google lists `gemini-3.8-live` as the replacement for both shutting-down models.
### Gemini Live 2.5 Flash
`gemini-2.5-flash-native-audio-preview-12-2025` · [Official docs](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-native-audio-preview-12-2025)
Provides "low-latency, real-time voice and video interactions with Gemini 2.5 Flash" through the Live API. It supports function calling, search grounding and thinking. Knowledge cutoff: January 2025.
Google lists a shutdown date of **November 17, 2026** for this preview model on the Gemini API, with Gemini Live 3.8 as the replacement.
| Metric | Value | Source |
| - | - | - |
| τ-Voice (agentic tasks) · **Third-party** | 22.8% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
### Gemini Live 3.1 Flash
`gemini-3.1-flash-live-preview` · [Official docs](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview)
"A legacy preview model for real-time dialogue and voice-first AI applications." It supports thinking levels (`minimal` by default, up to `high`) and synchronous function calling.
Google lists a shutdown date of **November 17, 2026** for this preview model on the Gemini API, and recommends updating to Gemini Live 3.8.
| Metric | Value | Source |
| - | - | - |
| Big Bench Audio (reasoning), high / minimal thinking · **Third-party** | 97% / 71% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| Time to first audio, high / minimal thinking · **Third-party** | 2.99 s / 0.96 s | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| τ-Voice (agentic tasks), high / minimal thinking · **Third-party** | 37.7% / 26.2% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
### Gemini Live 3.8
`gemini-3.8-live` · [Official docs](https://ai.google.dev/gemini-api/docs/models/gemini-3.8-live)
Google calls it the "default option for most low-latency voice agent experiences". It adds interleaved reasoning and asynchronous (non-blocking) function calling, and proactive audio is always on.
| Metric | Value | Source |
| - | - | - |
| Big Bench Audio (reasoning) · **Third-party** | 92% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| Time to first audio · **Third-party** | 1.18 s | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
| τ-Voice (agentic tasks) · **Third-party** | 30.1% | [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech) |
Google publishes its Gemini Live evaluations in a [model card](https://deepmind.google/models/model-cards/gemini-3-8-audio/) and an evaluation PDF but does not list benchmark scores on its model pages. The **Third-party** figures in this section are published by [Artificial Analysis](https://artificialanalysis.ai/speech-to-speech), an independent benchmarking company, not by Google. Accessed October 16, 2026.
***
Realtime model results depend heavily on settings such as reasoning effort or thinking level: higher settings usually improve accuracy but increase time to first audio. Benchmarks from different providers are not directly comparable. Test candidate models with real calls from your own use case.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.