Lexicost
AI voice agents / Costs, terms and calculators

OpenAI Realtime API cost per minute

Checked against its sources on

OpenAI Realtime API cost per minute is a figure OpenAI does not publish. Its realtime speech models are priced per million audio tokens, and the whole conversation so far is sent again as input on every turn, so later turns cost more than early ones and a long call costs more per minute than a short one.

Realtime speech model: what a call costs as it gets longer
Each total says what it is worked out from. A figure with a dotted line opens its source.

The formula

Audio cost of a call = caller tokens × input price + agent tokens × output price + re-sent tokens × re-sent price, where re-sent tokens = (caller tokens per turn + agent tokens per turn) × turns × (turns - 1) ÷ 2
  • Turns: minutes of the call times the agent's replies per minute, rounded, at least one.
  • Caller tokens per turn: call seconds × the caller's share of the call × tokens per second of user audio (from OpenAI's stated rate), divided by turns.
  • Agent tokens per turn: the same with the agent's share and the rate for assistant audio.
  • Re-sent price: the input price, with the share you expect to be cached priced at the cached input rate.

What OpenAI publishes for a realtime call

For gpt-realtime-2.1, the model the calculator opens on, OpenAI's audio prices are $32.00 per 1M audio input tokens (gpt-realtime-2.1), $0.40 per 1M audio cached input tokens (gpt-realtime-2.1) and $64.00 per 1M audio output tokens (gpt-realtime-2.1). The cost guide counts speech at 10 tokens per second of caller audio (1 per 100 ms) and 20 tokens per second of agent audio (1 per 50 ms). The mini model costs $10.00 per 1M audio input tokens (gpt-realtime-2.1-mini) and $20.00 per 1M audio output tokens (gpt-realtime-2.1-mini). Data residency costs more: $35.20 per 1M audio input tokens, data residency (gpt-realtime-2.1).

The OpenAI pages this site read give no cost per minute for these models. The cost guide says instead: Realtime token costs are hard to estimate ahead; OpenAI suggests measuring a sample session. Part of what makes it hard is a rule on the same page: The whole conversation is sent as input on every Response, so later turns cost more (Realtime API). Two of the older models still on the price list, at $32.00 per 1M audio input tokens (gpt-realtime) and $10.00 per 1M audio input tokens (gpt-realtime-mini), are listed as deprecated on OpenAI's deprecations page, with a shutdown date of Jan 20, 2027.

Why each turn costs more than the one before

OpenAI's cost guide describes a session as a series of turns, each triggering a Response: Charged per Response on input and output tokens; no charge for bandwidth or connections (Realtime API). The input is everything said so far. On the second turn the model hears the first exchange again, both sides of it, before the caller's new words; on the twentieth it hears nineteen earlier exchanges again.

As a call goes on at the same pace, the re-sent audio grows with the square of the number of turns: twice the length means twice the turns and close to four times the re-sent tokens. Per minute, that part of the bill roughly doubles when the call doubles in length. The caller's speech heard once and the agent's speech cost the same per minute at any length. The square holds only until the conversation outgrows the model's input limit: When a conversation exceeds the model's input limit, the oldest messages are dropped from the input (Realtime API).

Caching softens this. Re-sent audio that matches an earlier Response can be billed at the cached rate: Prompt caching is applied automatically to repeated input in multi-turn sessions, best-effort and not guaranteed (Realtime API). Developers can also shrink what is re-sent. Clients can set a smaller token window than the model maximum to control token use and cost (Realtime API). That works against caching: Truncation breaks the cache near the start of the conversation; truncating every turn keeps the cache rate low (Realtime API). The bill can be capped for the month: A hard spend limit makes affected API requests return a 429 error at the configured monthly amount.

How the calculator works it out, and what it leaves out

The prices, the token rates and the re-sending rule in the calculator are OpenAI's. Two simplifications are this site's own: each agent reply is one turn and one Response, and the call's speech is split evenly between the turns. Every other setting is your assumption, including the share of re-sent audio billed at the cached rate, which starts at 0%, the worst case. The bill comes out in the three items worked through below, followed by the same sum for a 1-minute call. The calculator counts no tokens for silence; OpenAI states this for the caller's side only: With voice activity detection on, empty audio is not counted as input tokens unless the client adds it (Realtime API).

It counts audio only. Text tokens for instructions and tools are billed on top, at $4.00 per 1M text input tokens (gpt-realtime-2.1). Input transcription, if enabled, is billed separately on its own model rate (Realtime API). For the phone line OpenAI points to a carrier: To connect a phone number, use a SIP trunking provider such as Twilio (Realtime API). The OpenAI pages this site read list no SIP charge of OpenAI's own (not publicly listed). The calculator also assumes the conversation window never fills, so on a call long enough to be truncated its re-sent item is too high.

A worked example, at the calculator's starting settings: gpt-realtime-2.1, a 5-minute call, 4 replies a minute, the caller speaking 40% of the time and the agent 40%, nothing cached. That makes 20 turns over 300 seconds. Caller tokens per turn: 300 seconds × 40% × 10 tokens per second of caller audio (1 per 100 ms), divided by 20 turns, is 60. Agent tokens per turn: 300 seconds × 40% × 20 tokens per second of agent audio (1 per 50 ms), divided by 20 turns, is 120.

The agent's speech is 20 × 120 = 2,400 tokens at $64.00 per 1M audio output tokens (gpt-realtime-2.1): $0.1536. The caller's speech heard once is 20 × 60 = 1,200 tokens at $32.00 per 1M audio input tokens (gpt-realtime-2.1): $0.0384. The re-sent conversation is (60 + 120) × 20 × 19 ÷ 2 = 34,200 tokens at the same input price: $1.0944. The call's audio comes to $1.2864, or $0.257 a minute; the calculator rounds the three items to $0.154, $0.0384 and $1.09 and the total to $1.29. The same settings give $0.073 a minute on a 1-minute call (4 turns, 1,080 re-sent tokens) and $0.488 a minute on a 10-minute call, $4.88 in all.

If, for example, 80% of the re-sent audio on the 5-minute call were billed at $0.40 per 1M audio cached input tokens (gpt-realtime-2.1) and the rest at the input price, the re-sent item would fall to $0.23 and the call to $0.422, or $0.0844 a minute.

Where realtime voice is sold by the minute

OpenAI sells voice sessions by the minute with GPT-Live-1, through a separate Live API: $0.05 per minute, billed per second (gpt-live-1); backend model and tools billed separately. Its minute is the session clock: GPT-Live billed time includes silence and time the backend is working. The number of calls at once is capped (gpt-live-1 concurrent sessions by usage tier: Free not supported, Build 50, Launch 300, Grow 500).

Retell prices OpenAI's realtime models by the minute: $0.38 per minute, GPT Realtime 2.1; $0.07 per minute, GPT Realtime 2.1 mini; $0.05 per minute, GPT-Live 1, plus the backend model rate. The pages this site read do not say whether Retell adds its $0.055 per minute (Retell Voice Infra) to a speech-to-speech call: the pricing page lists that fee on its Default tab, its Speech-to-Speech tab has no such line, and Retell's billing guide does not say. Either way, Retell's flat per-minute rates do not line up exactly with the calculator's audio-only figure, which changes with the length of the call.

Ultravox charges $0.05 per minute after 30 free minutes for calls on its own speech model (Speech understanding and generation included). ElevenLabs bills its Eleven Realtime models both ways at once (Realtime model: call minute plus tokens): the agent's call minute ($0.08 per minute, pay as you go) plus tokens at the Realtime model's rates, such as Eleven Realtime v1 Max: $1.80 per million input tokens.

Often confused with

Vendors cited on this page

ElevenLabs Agents (ElevenAgents), OpenAI, Retell AI, Ultravox. Each profile lists everything that vendor publishes, with its sources. All of them are side by side in the comparison.

Sources

Checked against its sources on 14 Nov 2026. Vendor figures on this page open their source when clicked. Lexicost has no commercial relationship with any vendor named on this site.

AI voice agent cost per minuteAdds up the five parts of a minute from each vendor's own published rates, with a box for any part a vendor does not price, and shows which parts each vendor's headline price leaves out.Billed minutes: how vendors round a callPuts seven vendors' own rules for rounding, minimums, attempt fees, silence and voicemail side by side, including where a vendor's pages disagree or say nothing, and its calculator works out how many minutes the same calls are billed at every vendor whose per-minute rates it holds.Concurrency limits for AI voice agentsPuts what vendors' own pages say about included calls at once, the price of one more and what happens to a call over the limit side by side, and works out how many calls at once a given busy hour needs.Cost per call: AI agent or a personWorks out an AI call's cost from a vendor's own published rate and a person's from Bureau of Labor Statistics wage and benefit figures, finds the finish rate at which the AI stops saving money, and puts side by side how vendors bill a call after the AI hands it to a person.Endpointing, turn detection and barge-inLines up what Retell, Vapi, Bland, ElevenLabs and Synthflow call the settings that decide when their agent starts and stops talking, with the defaults the vendors state and the places where a vendor's own pages disagree.
Source for every vendor figure on this page (47)
  1. OpenAI, realtime speech model rate: $32.00 per 1M audio input tokens (gpt-realtime-2.1) (gpt-realtime-2.1. pay as you go. Standard processing (not Batch, not data residency; the business pricing page lists those rates as separate facts); audio input tokens; Realtime API).gpt-realtime-2.1 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  2. OpenAI, realtime speech model rate: $0.40 per 1M audio cached input tokens (gpt-realtime-2.1) (gpt-realtime-2.1. pay as you go. Standard processing (not Batch, not data residency; the business pricing page lists those rates as separate facts); audio cached input tokens; Realtime API).gpt-realtime-2.1 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  3. OpenAI, realtime speech model rate: $64.00 per 1M audio output tokens (gpt-realtime-2.1) (gpt-realtime-2.1. pay as you go. Standard processing (not Batch, not data residency; the business pricing page lists those rates as separate facts); audio output tokens; Realtime API).gpt-realtime-2.1 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  4. OpenAI, audio tokens per second: 10 tokens per second of caller audio (1 per 100 ms) (Realtime API. "Conversational Realtime API sessions" (Cost optimization page, Realtime API tab, Per-Response costs); user (caller) audio; the sentence names no model).Audio tokens in user messages are 1 token per 100 ms of audio, while audio tokens in assistant messages are 1 token per 50ms of audio. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  5. OpenAI, audio tokens per second: 20 tokens per second of agent audio (1 per 50 ms) (Realtime API. "Conversational Realtime API sessions" (Cost optimization page, Realtime API tab, Per-Response costs); assistant (agent) audio; the sentence names no model).Audio tokens in user messages are 1 token per 100 ms of audio, while audio tokens in assistant messages are 1 token per 50ms of audio. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  6. OpenAI, realtime speech model rate: $10.00 per 1M audio input tokens (gpt-realtime-2.1-mini) (gpt-realtime-2.1-mini. pay as you go. Standard processing (not Batch, not data residency; the business pricing page lists those rates as separate facts); audio input tokens; Realtime API).gpt-realtime-2.1-mini Audio $10.00 $20.00 $0.30 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  7. OpenAI, realtime speech model rate: $20.00 per 1M audio output tokens (gpt-realtime-2.1-mini) (gpt-realtime-2.1-mini. pay as you go. Standard processing (not Batch, not data residency; the business pricing page lists those rates as separate facts); audio output tokens; Realtime API).gpt-realtime-2.1-mini Audio $10.00 $20.00 $0.30 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  8. OpenAI, realtime speech model rate: $35.20 per 1M audio input tokens, data residency (gpt-realtime-2.1) (gpt-realtime-2.1. pay as you go. "Data residency +10%" processing mode selected for Voice models on the business pricing page; audio input tokens).Audio: $35.20 / 1M tokens for inputs $0.44 / 1M tokens for cached inputs $70.40 / 1M tokens for outputs openai.com/business/pricing Read 14 Nov 2026.
  9. OpenAI, billing rule: Realtime token costs are hard to estimate ahead; OpenAI suggests measuring a sample session (Realtime API token-priced models).Given the complexity in Realtime API token usage it can be difficult to estimate your costs ahead of time. A good approach is to use the Realtime Playground with your intended prompts and functions, and measure the token usage over a sample session. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  10. OpenAI, billing rule: The whole conversation is sent as input on every Response, so later turns cost more (Realtime API) (Realtime API. "Conversational Realtime API sessions" (Cost optimization page, Realtime API tab)).The entire conversation is sent to the model for each Response. The output from a turn will be added as Items to the server Conversation and become the input to subsequent turns, thus turns later in the session will be more expensive. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  11. OpenAI, realtime speech model rate: $32.00 per 1M audio input tokens (gpt-realtime) (gpt-realtime. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio input tokens; Realtime API).gpt-realtime Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  12. OpenAI, realtime speech model rate: $10.00 per 1M audio input tokens (gpt-realtime-mini) (gpt-realtime-mini. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio input tokens; Realtime API).gpt-realtime-mini Audio $10.00 $20.00 $0.30 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  13. OpenAI, billing rule: Charged per Response on input and output tokens; no charge for bandwidth or connections (Realtime API) (Realtime API speech-to-speech sessions).Realtime API costs are accrued when a Response is created, and is charged based on the numbers of input and output tokens (except for input transcription costs, see below). There is no cost currently for network bandwidth or connections. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  14. OpenAI, billing rule: When a conversation exceeds the model's input limit, the oldest messages are dropped from the input (Realtime API) (Realtime API speech-to-speech sessions).When the number of tokens in a conversation exceeds the model’s input token limit the conversation be truncated, meaning messages (starting from the oldest) will be dropped from the Response input. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  15. OpenAI, billing rule: Prompt caching is applied automatically to repeated input in multi-turn sessions, best-effort and not guaranteed (Realtime API) (Realtime API speech-to-speech sessions).Realtime API supports prompt caching, which is applied automatically and can dramatically reduce the costs of input tokens during multi-turn sessions. Caching applies when the input tokens of a Response match tokens from a previous Response, though this is best-effort and not guaranteed. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  16. OpenAI, billing rule: Clients can set a smaller token window than the model maximum to control token use and cost (Realtime API) (Realtime API speech-to-speech sessions).Clients can set a smaller token window than the model’s maximum, which is a good way to control token usage and cost. This is controlled with the token_limits.post_instructions configuration (if you configure truncation with a retention_ratio type as shown below). developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  17. OpenAI, billing rule: Truncation breaks the cache near the start of the conversation; truncating every turn keeps the cache rate low (Realtime API) (Realtime API speech-to-speech sessions).Truncation busts the cache near the beginning of the conversation, and if truncation occurs on every turn then cache rate will be very low. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  18. OpenAI, billing rule: A hard spend limit makes affected API requests return a 429 error at the configured monthly amount (organization or project setting).Hard spend limit Affected API requests return a 429 error Enforce a monthly organization or project cap developers.openai.com/api/docs/guides/rate-limits Read 14 Nov 2026.
  19. OpenAI, billing rule: With voice activity detection on, empty audio is not counted as input tokens unless the client adds it (Realtime API) (Realtime API with voice activity detection (VAD) turned on).VAD will effectively filter out empty input audio, so empty audio doesn’t count as input tokens unless the client manually adds it as conversation input. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  20. OpenAI, aI model rate per million tokens: $4.00 per 1M text input tokens (gpt-realtime-2.1) (gpt-realtime-2.1. pay as you go. Standard processing (not Batch, not data residency; the business pricing page lists those rates as separate facts); text input tokens; Realtime API).gpt-realtime-2.1 Audio $32.00 $64.00 $0.40 Text $4.00 $24.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  21. OpenAI, billing rule: Input transcription, if enabled, is billed separately on its own model rate (Realtime API) (Realtime API sessions with input transcription enabled).If enabled, input transcription is billed separately by the Realtime API. Input transcription uses a different model and rate card from the speech-to-speech model. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  22. OpenAI, other phone rates: To connect a phone number, use a SIP trunking provider such as Twilio (Realtime API) (Realtime API over SIP).If you want to connect a phone number to the Realtime API, use a SIP trunking provider (e.g., Twilio). This is a service that converts your phone call to IP traffic. developers.openai.com/api/docs/guides/voice-sip Read 14 Nov 2026.
  23. OpenAI, other phone rates: Not publicly listed (Realtime API and GPT-Live over SIP). Read 14 Nov 2026.
  24. OpenAI, realtime speech model rate: $0.05 per minute, billed per second (gpt-live-1); backend model and tools billed separately (gpt-live-1. pay as you go. Standard processing (not Batch, not data residency); voice session duration; Live API (/v1/live/sessions)).Voice sessions cost $0.05 per minute, billed per second. Backend model and tool usage is billed separately. developers.openai.com/api/docs/models/gpt-live-1 Read 14 Nov 2026.
  25. OpenAI, billing rule: GPT-Live billed time includes silence and time the backend is working (gpt-live-1, Live API).Active session time includes time when the user speaks, the assistant speaks, both are silent, or the backend is working. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  26. OpenAI, other limits: gpt-live-1 concurrent sessions by usage tier: Free not supported, Build 50, Launch 300, Grow 500 (gpt-live-1, per usage tier).Rate limits are measured in concurrent sessions. TIER CONCURRENT SESSIONS Free Not supported Build 50 Launch 300 Grow 500 developers.openai.com/api/docs/models/gpt-live-1 Read 14 Nov 2026.
  27. Retell AI, realtime speech model rate: $0.38 per minute, GPT Realtime 2.1 (Pay as you go. Speech-to-speech LLM).GPT Realtime 2.1: $0.38/minute retellai.com/pricing Read 14 Nov 2026.
  28. Retell AI, realtime speech model rate: $0.07 per minute, GPT Realtime 2.1 mini (Pay as you go. Speech-to-speech LLM).GPT Realtime 2.1 mini: $0.07/minute retellai.com/pricing Read 14 Nov 2026.
  29. Retell AI, realtime speech model rate: $0.05 per minute, GPT-Live 1, plus the backend model rate (Pay as you go. Speech-to-speech LLM).GPT-Live 1: + backend LLM price per minute, $0.05/minute retellai.com/pricing Read 14 Nov 2026.
  30. Retell AI, platform fee per minute: $0.055 per minute (Retell Voice Infra) (Pay as you go. Voice (call) agents; listed as "Retell Voice Infra" under "Conversation Voice Engine" in the Default tab).Retell Voice Infra: $0.055/minute retellai.com/pricing Read 14 Nov 2026.
  31. Ultravox, platform fee per minute: $0.05 per minute after 30 free minutes (Pay As You Go. pay as you go. after the first 30 free minutes).First 30 minutes free, $0.05/min after that. ultravox.ai/pricing Read 14 Nov 2026.
  32. Ultravox, what that fee includes: Speech understanding and generation included (Ultravox v0.7 model on Ultravox Realtime).We’re also keeping the cost of using v0.7 on Ultravox Realtime (our platform for building and scaling voice AI agents) the same at $.05/min. That’s state-of-the-art model performance with speech understanding and speech generation (including ElevenLabs and Cartesia voices) for a price that is half the cost of most providers. ultravox.ai/blog/introducing-ultravox-v0-7-the-world-s-smartest-speech-understanding-model Read 14 Nov 2026.
  33. ElevenLabs Agents (ElevenAgents), how the AI model is charged: Realtime model: call minute plus tokens (Voice agent using an Eleven Realtime model instead of a separate AI model).A Realtime agent is billed the standard ElevenAgents per-minute rate, plus its input, cached input, and output tokens at the Realtime model’s rates. elevenlabs.io/docs/eleven-agents/customization/realtime-models Read 14 Nov 2026.
  34. ElevenLabs Agents (ElevenAgents), platform fee per minute: $0.08 per minute, pay as you go (Pay as you go. pay as you go. Cost calculator with "Pay as you go" selected; LLM shown as a separate line).Platform voice ($0.08/min) elevenlabs.io/pricing/agents Read 14 Nov 2026.
  35. ElevenLabs Agents (ElevenAgents), aI model rate per million tokens: Eleven Realtime v1 Max: $1.80 per million input tokens (Eleven Realtime v1 Max, input tokens).Eleven Realtime v1 Max, Input tokens (per 1M): $1.80 elevenlabs.io/pricing/api Read 14 Nov 2026.
  36. OpenAI, billing rule: GPT-Live sessions are billed per second, with no rounding up to a whole minute (gpt-live-1, Live API).GPT-Live voice sessions are billed per second at the current model rate. Session duration is not rounded up to the next whole minute. developers.openai.com/api/docs/guides/voice-latency-cost Read 14 Nov 2026.
  37. OpenAI, realtime speech model rate: $0.30 per 1M audio cached input tokens (gpt-realtime-2.1-mini) (gpt-realtime-2.1-mini. pay as you go. Standard processing (not Batch, not data residency; the business pricing page lists those rates as separate facts); audio cached input tokens; Realtime API).gpt-realtime-2.1-mini Audio $10.00 $20.00 $0.30 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  38. OpenAI, realtime speech model rate: $32.00 per 1M audio input tokens (gpt-realtime-2) (gpt-realtime-2. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio input tokens; Realtime API).gpt-realtime-2 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  39. OpenAI, realtime speech model rate: $0.40 per 1M audio cached input tokens (gpt-realtime-2) (gpt-realtime-2. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio cached input tokens; Realtime API).gpt-realtime-2 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  40. OpenAI, realtime speech model rate: $64.00 per 1M audio output tokens (gpt-realtime-2) (gpt-realtime-2. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio output tokens; Realtime API).gpt-realtime-2 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  41. OpenAI, realtime speech model rate: $32.00 per 1M audio input tokens (gpt-realtime-1.5) (gpt-realtime-1.5. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio input tokens; Realtime API).gpt-realtime-1.5 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  42. OpenAI, realtime speech model rate: $0.40 per 1M audio cached input tokens (gpt-realtime-1.5) (gpt-realtime-1.5. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio cached input tokens; Realtime API).gpt-realtime-1.5 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  43. OpenAI, realtime speech model rate: $64.00 per 1M audio output tokens (gpt-realtime-1.5) (gpt-realtime-1.5. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio output tokens; Realtime API).gpt-realtime-1.5 Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  44. OpenAI, realtime speech model rate: $0.30 per 1M audio cached input tokens (gpt-realtime-mini) (gpt-realtime-mini. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio cached input tokens; Realtime API).gpt-realtime-mini Audio $10.00 $20.00 $0.30 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  45. OpenAI, realtime speech model rate: $20.00 per 1M audio output tokens (gpt-realtime-mini) (gpt-realtime-mini. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio output tokens; Realtime API).gpt-realtime-mini Audio $10.00 $20.00 $0.30 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  46. OpenAI, realtime speech model rate: $0.40 per 1M audio cached input tokens (gpt-realtime) (gpt-realtime. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio cached input tokens; Realtime API).gpt-realtime Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.
  47. OpenAI, realtime speech model rate: $64.00 per 1M audio output tokens (gpt-realtime) (gpt-realtime. pay as you go. Rate as listed on the developer pricing page, which shows no Batch or data-residency rate for this model; audio output tokens; Realtime API).gpt-realtime Audio $32.00 $64.00 $0.40 developers.openai.com/api/docs/pricing Read 14 Nov 2026.