What is GPT-Realtime-2.1?
GPT-Realtime-2.1 is OpenAI's current Realtime voice model, an update to GPT-Realtime-2 with the same price and limits. One model listens, reasons as much as you allow, calls functions or remote MCP tools and replies in speech, with no separate transcription or text-to-speech step. OpenAI says it recognises spoken letters and numbers better and copes better with silence, background noise and interruptions. Its knowledge stops at September 2024.
OpenAI names it the replacement for its retiring older Realtime models. For new conversational apps, though, OpenAI now points first to GPT-Live, which keeps talking while a separate backend does the reasoning.
What changed from GPT-Realtime-2?
It hears the details better: OpenAI improved how it handles spoken letters and numbers, silence and noise, and interruptions, and kept everything else the same.
- Letters and numbers
- Better recognition of spelled names, order numbers, codes and other alphanumeric values.
- Silence and noise
- Better at not reacting to silence and background noise on the line.
- Interruptions
- Better behaviour when the caller talks over it.
- Price and limits
- Unchanged: $32 / $64 per million audio tokens, a 128K context, 32K output tokens and knowledge up to September 2024.
Who is GPT-Realtime-2.1 best for?
GPT-Realtime-2.1 is best for phone and support agents that must catch numbers and names, voice agents that call tools, voice apps that take images, and apps moving off older Realtime models.
Phone and support agents
Callers read out order numbers, spell names and talk over background noise, which is what the 2.1 update was built to handle.
Voice agents that call tools
One model hears the request, decides what to do, calls your functions or a remote MCP server and answers aloud in the same session.
Voice apps that use images
Callers can share a photo or screenshot mid-session, which GPT-Live-1 cannot take directly.
Apps on older Realtime models
OpenAI names it the replacement for the retiring gpt-realtime and gpt-4o-realtime models, on the same Realtime API.
What's included with GPT-Realtime-2.1?
Adjustable reasoning, spoken preambles, function and MCP tools, image input, browser, server and phone connections, automatic prompt caching and a cheaper mini version.
- Adjustable reasoning
- You set how much it thinks before it speaks; OpenAI suggests starting low for most voice agents, since higher effort adds delay and output tokens.
- Spoken preambles
- It can say a short line such as "let me check" while it reasons or calls a tool, kept apart from its final answer.
- Function and MCP tools
- Your app runs its own functions, or the Realtime API calls a remote MCP server for you.
- Image input
- Images go in next to audio and text, at $5 per million tokens.
- Browser, server and phone connections
- WebRTC for browsers, WebSockets for servers and SIP for phone lines, plus the Agents SDK's RealtimeAgent for a quick start.
- Automatic prompt caching
- Repeated context bills at $0.40 per million tokens, which matters because every reply re-sends the conversation.
- A cheaper mini version
- GPT-Realtime-2.1 mini costs $10 / $20 per million audio tokens, though OpenAI says it is weaker at following instructions and calling functions.
How much does GPT-Realtime-2.1 cost?
GPT-Realtime-2.1 costs $32 per million audio input tokens and $64 per million audio output tokens on the OpenAI API, with text at $4 / $24, image input at $5 and cached input at $0.40.
- Cached text input $0.40 per million tokens
- Cached image input $0.50 per million tokens
- Images go in only; replies are audio or text
- Cached audio input $0.40 per million tokens
- Caller speech counts 1 token per 100 ms; model speech 1 per 50 ms
GPT-Realtime-2.1 runs only on the Realtime API and is not on the free API tier. Each reply re-sends the whole conversation as input, so later turns cost more, higher reasoning effort adds output tokens, and input transcription is billed separately. GPT-Realtime-2.1 mini costs $10 / $20 per million audio tokens and $0.60 / $2.40 for text. GPT-Realtime-2 is priced the same as 2.1. Full pricing on developers.openai.com. We checked these prices on .
How do readers rate GPT-Realtime-2.1?
No one has rated GPT-Realtime-2.1 yet. Be the first.
Sign in to rate. One rating per account.
GPT-Realtime-2.1 pros and cons
It handles tools, numbers and noisy calls well in one model, but costs grow with every turn and OpenAI now points new conversational apps to GPT-Live.
Pros
- Improves the spots where voice agents often slip, letters and numbers, silence and noise, and interruptions, at the same price as GPT-Realtime-2.
- Developers who moved to GPT-Live say Realtime was steadier for strict phone flows, where a confirmed answer has to fire the right tool every time.
- Testers of the Realtime 2 models call them impressive with tools and reasoning switched on, handling loaded context and follow-up questions without sounding generic.
- One model listens, reasons, calls tools and speaks, and it takes images, which GPT-Live-1 cannot.
Cons
- Costs climb fast and are hard to predict: every reply re-sends the conversation, and one developer testing 2.1 measured about $5 a day for a very active user.
- With tool choice left on auto, testers say it calls tools it does not need; forcing a required tool cut one developer's bill by about 30%.
- The cheaper 2.1 mini is weaker at tools, OpenAI says, and one tester found it failed most of their tool-heavy use cases, so busy agents pay the full rate.
- Its knowledge stops at September 2024, a session ends after 60 minutes, there is no free API tier, and OpenAI now steers new conversational apps to GPT-Live.
What are the best GPT-Realtime-2.1 alternatives?
Four tools on Recaply do GPT-Realtime-2.1's job. Chroma 1.0 is open source, so you can run it on your own server.
GPT-Live-1PaidChoose it when natural turn-taking matters more than one-model simplicity; it bills a flat $0.05 a minute plus your backend, and OpenAI now recommends it for new voice apps.
Gemini 3.8 LiveFreeChoose it when price matters most: its paid audio costs $3 / $12 per million tokens against $32 / $64 here, it has a free tier, and it also takes video.
Grok Voice Think Fast 2.0PaidChoose it when you want a flat $0.08 a minute instead of token billing, with web search, X search and MCP tools built in.
Chroma 1.0Open SourceChoose it when you want a speech-to-speech dialogue model with voice cloning that you can host yourself under Apache 2.0.
Questions about GPT-Realtime-2.1
What does a minute of GPT-Realtime-2.1 cost?
There is no flat rate. A minute of caller speech is about 600 audio tokens and a minute of model speech about 1,200, so fresh audio costs roughly 2 cents a minute in and 8 cents out. Every reply re-sends the conversation, so long calls cost more unless the $0.40 cached rate applies.
Who is GPT-Realtime-2.1 for?
GPT-Realtime-2.1 is best for phone and support agents that must catch numbers and names, voice agents that call tools, voice apps that take images, and apps moving off older Realtime models.
Is GPT-Realtime-2.1 free?
No. GPT-Realtime-2.1 is a paid tool, from $4 / $24 per million input / output tokens.
How much does GPT-Realtime-2.1 cost?
GPT-Realtime-2.1's paid plans are Text at $4 / $24 per million input / output tokens, Image input at $5 per million input tokens and Audio at $32 / $64 per million input / output tokens.
Is there an open source alternative to GPT-Realtime-2.1?
Yes. Chroma 1.0 is an open source tool that does the same job, so you can run your own.
What are the best GPT-Realtime-2.1 alternatives?
The closest alternatives on Recaply are GPT-Live-1, Gemini 3.8 Live, Grok Voice Think Fast 2.0 and Chroma 1.0.
Where can I try GPT-Realtime-2.1?
On its official website, developers.openai.com.
Should I use GPT-Realtime-2.1 or GPT-Live-1?
Use GPT-Realtime-2.1 when one model should handle speech, reasoning and tool calls in a single session, or when callers share images. OpenAI points new conversational apps to GPT-Live-1, which listens while it speaks, hands the work to a separate backend and bills $0.05 a minute.
What is GPT-Realtime-2.1 mini?
It is a smaller, faster version of GPT-Realtime-2.1 at $10 / $20 per million audio tokens, with the same 128K context. OpenAI says the mini is weaker at following instructions and calling functions, and suggests building on the full model first, then testing whether mini is enough.
Does GPT-Realtime-2.1 work on phone calls?
Yes. The Realtime API accepts phone calls over SIP from your telephony provider, alongside WebRTC for browsers and WebSockets for servers, and OpenAI's SIP guide uses gpt-realtime-2.1 in its examples.
Our verdict
GPT-Realtime-2.1 is OpenAI's best choice for a voice agent where one model must listen, think and call tools.
It keeps GPT-Realtime-2's reasoning and tool use at the same price, handles numbers, noise and interruptions better, and takes images, which GPT-Live-1 cannot.
- Skip it for
- Simple conversational voice apps, where OpenAI now recommends GPT-Live-1 and its flat $0.05 a minute.
- Check first
- Model your cost on real call lengths, since every reply re-sends the conversation and the bill grows with each turn.
Eleven Realtime
GPT-Realtime-2
Scribe v2 Realtime
Voxtral 3
Gemini 3.8 Flash-Lite TTS