gpt-realtime
$32 / $64per 1M
Audio in / out
Text $4 / $16
cached $0.40
mini
$10 / $20per 1M
Audio input / output
Text $0.60 / $2.40
cached $0.30
Flagship / min
6 to 11cest
Token to minute
Third party
estimate
Mini / min
2 to 5cest
Converted from tokens
Third party
estimate
Per-minute figures are third-party token-to-minute estimates, not OpenAI rates. OpenAI leaves out telephony, and the Realtime API ships as a model rather than a phone agent framework.
OpenAI Realtime API plans compared: tokens, models, and rates
The Realtime API is a speech-to-speech model billed per token, not per minute. Each model and modality carries its own rate below.
The flagship realtime model, with cached audio input at $0.40 per million, for the highest quality voice.
Text tokens on the flagship, for the parts of a session handled as text rather than audio.
The cheaper realtime model, trading some quality for a lower rate, with cached audio at $0.30.
gpt-4o-transcribe for speech to text, or $0.003 on mini, billed per token.
tts-1 and tts-1-hd for non realtime text to speech, separate from the Realtime API.
Rates checked on OpenAI's pricing page on 21 October 2026. OpenAI lists no per-minute rate: a voice minute depends on how much the model listens and speaks, so any per-minute figure is an estimate.
How much the OpenAI Realtime API really costs per minute
Because it bills per token, the real cost of a call depends on speech length, context size, and caching. Four pieces make up an OpenAI Realtime minute.
Which OpenAI Realtime plan fits your use case
The flagship model is the pick when voice quality and reasoning matter most, at the higher token rate, so it suits agents where a natural, capable voice is the product rather than a cost line to minimize.
The mini model is cheaper for high volume or simpler calls, at a lower per-minute estimate, so routing routine turns to it and reserving the flagship for hard moments is the main lever on the bill.
Cached input cuts the cost of the same system prompt or context repeating across turns, which matters for long agent sessions where re-sending the setup every turn would otherwise dominate the token count.
The Realtime API has no telephony of its own, so a carrier or SIP provider, billed per minute on top of the token cost, is yours to add, and that layer belongs in any honest per-minute comparison.
Volume commitments and custom terms come through OpenAI sales, for teams whose usage or compliance needs go beyond the pay-as-you-go token rates the public API charges.
OpenAI Realtime API pricing calculator: estimate your monthly bill
The OpenAI figure uses a third-party token-to-minute estimate, about $0.09 a minute on gpt-realtime and $0.035 on mini, before telephony. The PyAI figure is your minutes at $0.05/min + $0.01 telephony, the Omni rate with the line included. OpenAI mini can cost less per minute, while PyAI is a fixed per-minute price.
OpenAI Realtime hidden costs: tokens, telephony, and orchestration
- Token pricing is unpredictable: a call's cost depends on how long the model speaks and listens, the context size, and caching, so a minute is not a fixed price.
- Output costs more than input: audio output at $64 per million is double the input, so a talkative agent costs more.
- Telephony is separate: the Realtime API stops at the model, so a carrier or SIP provider bills its own per-minute rate on top.
- You build the orchestration: OpenAI ships no phone agent framework, so call handling, barge in, and transfers fall to your own code.
- Caching helps only sometimes: cached input is far cheaper, but only when the same context repeats across turns.
- Transcription bills apart: speech to text with gpt-4o-transcribe is a separate per token charge.
- Text to speech is separate: tts-1 and tts-1-hd are priced per character, apart from the Realtime API.
OpenAI Realtime API reviews: what users say
We looked for real, attributed user reviews of OpenAI Realtime API that mention pricing and the product. As of October 2026 we found no public user reviews on the sources below, so there is nothing to quote honestly. The OpenAI Realtime API is a newer speech-to-speech developer product with no public review base of its own. The OpenAI listing on G2 is for Whisper transcription, a different product, so there are no realtime-voice reviews to quote without mixing the two.
No public user reviews found
Evaluate OpenAI Realtime API on a short trial and a reference call rather than on ratings, since no verifiable user reviews exist yet.
OpenAI Realtime API vs PyAI: price and billing compared
| OpenAI Realtime API | PyAI | |
|---|---|---|
| Billing model | Per token | ✓ Per call minute, no token meter |
| Voice rate | $0.06 to $0.11/min flagship, about $0.02 to $0.05 mini | Omni $0.05/min + $0.01 telephony, Agents $0.08 plus the line |
| Predictability | Varies with tokens and context | ✓ Same cost for every call minute |
| Model quality | ✓ Frontier OpenAI models | Hosted voice, OpenAI compatible |
| Telephony | Not included | ✓ $0.01 telephony minute as its own line |
| Phone agent layer | You build it | ✓ Built in phone agent |
| Speech to text and TTS | Separate per token products | ✓ Hear $0.001 and Speak $0.04 a minute, one invoice |
| Free to start | Usage billed, no set credit | ✓ $50 credit, no credit card |
| Output formats | Audio streams | ✓ WAV, PCM, or G.711 for carriers |
| Reviews | No API review aggregate | No G2 profile yet |
Choose OpenAI Realtime if you want
- Frontier OpenAI model quality and reasoning
- The lowest per minute on the mini model for simple calls
- Full control to assemble telephony, turn taking, and tools around the model
Choose PyAI if you want
- A set price per call minute, not a bill that moves with OpenAI token counts
- Telephony and a built-in phone agent billed alongside the voice minutes
- Phone line audio in WAV, PCM, or G.711, where Realtime returns raw audio streams
OpenAI Realtime cost by volume: voice minutes a month
Since the Realtime API bills per token, a monthly estimate depends on the model and how talkative the agent is. The table below uses midpoint per-minute estimates.
| Use case | What you need | Model | Watch out for |
|---|---|---|---|
| Highest quality | Complex conversations | gpt-realtime | Higher token cost |
| Cost sensitive | Simple, high volume calls | gpt-realtime-mini | Lower quality on hard tasks |
| Phone calls | Telephony and orchestration | Add your own stack | Billed on top of tokens |
OpenAI Realtime vs PyAI monthly cost by volume
OpenAI uses midpoint per-minute estimates. The PyAI Omni column prices every minute in every row at $0.05/min + $0.01 telephony. OpenAI mini can run lower per minute, while PyAI is a fixed, predictable price.
| Voice minutes a month | gpt-realtime (est) | mini (est) | PyAI Omni |
|---|---|---|---|
| 1,000 | $90 | $35 | $60 |
| 10,000 | $900 | $350 | $600 |
| 50,000 | $4,500 | $1,750 | $3,000 |
| 100,000 | $9,000 | $3,500 | $6,000 |
OpenAI figures use third-party token-to-minute estimates, about $0.09 on gpt-realtime and $0.035 on mini, before telephony. PyAI Omni is priced at $0.05/min + $0.01 telephony, so each minute is $0.05 plus the $0.01 telephony line. OpenAI mini costs less per minute, while PyAI is a fixed price that is easier to forecast.
OpenAI Realtime API pricing: the bottom line
The OpenAI Realtime API is priced per token, at $32 and $64 per million audio tokens on gpt-realtime and $10 and $20 on mini, which third parties estimate at about $0.06 to $0.11 and $0.02 to $0.05 a voice minute. Realtime API buyers add their own carrier and call orchestration on top. PyAI bills per minute instead, with Omni at $0.05/min + $0.01 telephony and Agents at $0.08, a fixed price that bundles telephony and a phone agent, while OpenAI mini can cost less per minute. Sandbox keys are free, with $50 in credit and no credit card.
PyAI does not run OpenAI's frontier models and does not undercut gpt-realtime-mini on raw per-minute cost. PyAI sells voice infrastructure by the minute, so speech to text, text to speech, and a phone agent land on one invoice.
OpenAI Realtime API pricing FAQs
The questions teams ask most before choosing the OpenAI Realtime API.
How much does the OpenAI Realtime API cost?
OpenAI charges for the Realtime API by the token, not the minute. Audio on gpt-realtime runs $32 per million input tokens and $64 per million output, while gpt-realtime-mini audio is $10 and $20. Converted to time, third-party estimates place a voice minute near $0.06 to $0.11 on the flagship and $0.02 to $0.05 on mini.
What is the difference between gpt-realtime and the mini model?
The gpt-realtime model costs $32 per million audio input tokens and $64 per million output. The gpt-realtime-mini model is cheaper at $10 and $20, trading some quality for a lower rate. Cached audio input is $0.40 per million on the flagship and $0.30 on mini, which lowers cost when context repeats.
How much is an OpenAI Realtime voice minute?
OpenAI does not publish a per-minute rate, since the Realtime API bills per token. Third-party conversions put a voice minute at roughly $0.06 to $0.11 on gpt-realtime and $0.02 to $0.05 on gpt-realtime-mini, depending on how much the model speaks and listens. Caching and voice activity detection lower the effective cost.
Does the OpenAI Realtime API include telephony?
No. The Realtime API is a model API, so telephony is not included. You connect your own carrier or SIP provider and pay their per-minute rates on top of the token cost. The API also does not bundle a phone agent framework, so you build the call handling and orchestration yourself.
Why is OpenAI Realtime pricing hard to forecast?
Because it bills per token, the cost of a call depends on how long the model listens and speaks, the context size, and caching, which vary by conversation. A token-to-minute estimate is only a guide. Teams that want a fixed number budget with a per-minute provider instead, where a minute is a minute.
What is a more predictable alternative to the OpenAI Realtime API?
PyAI charges per call minute, not per audio token, so each minute has a set price. Omni speech-to-speech is $0.05/min + $0.01 telephony and Agents is $0.08 plus the same line, so the invoice stays predictable. OpenAI Realtime mini can cost less per minute on raw tokens, while PyAI is easier to forecast and bundles the stack.
More OpenAI voice comparisons and guides
Price PyAI against your own volume.
Per minute pricing with no plan fee and no credits to burn. Speak is $0.04/min, Cast $0.02/min, Hear $0.001/min, and Omni $0.05/min + $0.01 telephony, with $50 in starter credit and no credit card.
