gpt-realtime-2
Summary
gpt-realtime-2 is a proprietary language model from OpenAI, released on 7 May 2026. It accepts text and audio and generates text and audio, with a 128K-token context window and up to 32K output tokens per response. As of 21 Oct 2026, the lowest listed API price is $4 per million input tokens and $24 per million output tokens (Vercel AI Gateway), across 1 provider we track.
Realtime speech-to-speech model with configurable reasoning and reliable tool use for voice agents
Last verified 21 Oct 2026. Every figure on this page links to its source.
Specifications
| Developer | OpenAI |
|---|---|
| Release date | 7 May 2026 [models.dev] |
| Status | Generally available [models.dev] |
| Weights | Proprietary (API only) [models.dev] |
| Context window | 128K (128,000 tokens) [models.dev] |
| Max output | 32K tokens [models.dev] |
| Knowledge cutoff | 30 Sept 2024 [models.dev] |
| Input | Text and audio [models.dev] |
| Output | Text and audio [models.dev] |
| Reasoning mode | No [models.dev] |
| Tool calling | Yes [models.dev] |
gpt-realtime-2 API pricing by provider
USD per million tokens. Sorted by input price.
| Provider | Input | Output | Cached input | Context | Source |
|---|---|---|---|---|---|
| Vercel AI Gateway openai/gpt-realtime-2 | $4 | $24 | $0.40 | 128K | models.dev, 21 Oct 2026 |
Frequently asked questions
What is the context window of gpt-realtime-2?
gpt-realtime-2 has a context window of 128,000 tokens (128K) and can generate up to 32,000 tokens in a single response, according to models.dev.
How much does the gpt-realtime-2 API cost?
gpt-realtime-2 is listed from $4 per million input tokens and $24 per million output tokens (Vercel AI Gateway). Prices last verified 21 Oct 2026.
Is gpt-realtime-2 open source?
No. gpt-realtime-2 is a proprietary model; OpenAI has not released its weights. It is available through APIs including Vercel AI Gateway.
When was gpt-realtime-2 released?
OpenAI released gpt-realtime-2 on 7 May 2026, according to models.dev.
Change history
- Context window changed from none to 128K
- Max output changed from none to 32K
- Tool calling changed from no to yes
- Added to the directory
- Now listed on Vercel AI Gateway