OpenAI pricing
OpenAI sells voice through its general API. Transcription and text to speech are available as standalone models, and the realtime models handle live speech-to-speech conversations in a single connection.
Realtime billing is by audio tokens rather than minutes, so the cost per minute of conversation depends on how much each side talks. The calculator converts tokens to minutes with a fixed assumption, which is why the realtime row is marked as an estimate.
If you already use OpenAI models for the rest of your application, a single account and invoice can matter as much as a small price gap. Compare accuracy on your own audio before choosing on price alone.
Some “Sign up” links are affiliate links and may earn a commission. Rankings are computed from price only and never take affiliate status into account.
Every month I transcribe of audio.
AssemblyAI and Mistral Voxtral tie for cheapest, at $3.50 a month. Azure would cost 4.8 times as much.
Estimated monthly bill for 1,000 minutes, lowest first. 7 of 7 providers fit.
- 1AssemblyAIUniversal-3.5 Pro (pre-recorded), $0.21/hr$3.50 per monthLowest priceVisit AssemblyAI
Usage $3.50 Monthly total $3.50 Works out to $0.0035 per minute.
Priced in the "Above 0" tier; every unit is billed at that tier's rate.
- 1Mistral VoxtralVoxtral Mini Transcribe 3, $0.0035/min$3.50 per monthLowest priceVisit Voxtral
Usage $3.50 Monthly total $3.50 Works out to $0.0035 per minute.
Priced in the "Above 0" tier; every unit is billed at that tier's rate.
- 3ElevenLabsScribe v2 (and v2 Medical), $0.22/hr$3.67 per month+$0.17, 4.8% moreVisit ElevenLabs
Usage $3.67 Monthly total $3.67 Works out to $0.00367 per minute.
Priced in the "Above 0" tier; every unit is billed at that tier's rate.
- 4DeepgramNova-3 (pre-recorded), $0.0043/min$4.30 per month+$0.80, 22.9% moreVisit Deepgram
Usage $4.30 Monthly total $4.30 Works out to $0.0043 per minute.
Priced in the "Above 0" tier; every unit is billed at that tier's rate.
- 5OpenAIThis pageGPT Transcribe, $0.0045/min$4.50 per month+$1.00, 28.6% moreVisit OpenAI
Usage $4.50 Monthly total $4.50 Works out to $0.0045 per minute.
Priced in the "Above 0" tier; every unit is billed at that tier's rate.
- 6Google CloudSpeech-to-Text V2 Standard (incl. Chirp), $0.016/min$16.00 per month+$12.50, 357.1% moreVisit Google Cloud
Usage $16.00 Monthly total $16.00 Works out to $0.016 per minute.
Tiers are graduated: each band of usage is billed at its own rate (reached: Up to 500000).
- 7AzureStandard real-time transcription, $1/hr$16.67 per month+$13.17, 376.2% moreVisit Azure
Usage $16.67 Monthly total $16.67 Works out to $0.0167 per minute.
Priced in the "Above 0" tier; every unit is billed at that tier's rate.
- Prices updated 2026-10-30.
- Priced in the "Above 0" tier; every unit is billed at that tier's rate.
- Free-tier credits are not subtracted.
- How we calculate this
Some “Visit” links are affiliate links: the provider may pay us if you sign up. That never changes the order, which is set by price alone.
Pricing by model
GPT Transcribe pricing
Speech to text. Current recommended speech-to-text model for recorded files and committed realtime turns, with context, keyword and language hints. No speaker labels.
| Monthly usage | Price per minute |
|---|---|
| Any volume | $0.0045 |
gpt-transcribe is billed per minute of audio ($0.0045/min, listed on the pricing page and the model page). No speaker labels in this model; see diarization add-on. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled. EU data residency (eu.api.openai.com, EEA + Switzerland) is available but requires Modified Abuse Monitoring or Zero Data Retention and adds a 10% uplift for models released on or after 2026-03-05 that are eligible (not included in this price).
Source: developers.openai.com
GPT-4o Transcribe pricing
Speech to text. Earlier GPT-4o based speech-to-text model with better word error rate than Whisper. Deprecated, shuts down 2027-02-26.
| Monthly usage | Price per minute |
|---|---|
| Any volume | $0.006 |
Token-billed ($2.50 input / $10.00 output per 1M tokens); OpenAI quotes an estimated $0.006/min, used here. Deprecated: OpenAI announced this model will shut down on 2027-02-26 (https://developers.openai.com/api/docs/guides/speech-to-text); price remains valid until then. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled.
Source: developers.openai.com
GPT-4o Mini Transcribe pricing
Speech to text. Lower-cost GPT-4o mini based speech-to-text model. Deprecated, shuts down 2027-02-26.
| Monthly usage | Price per minute |
|---|---|
| Any volume | $0.003 |
Token-billed ($1.25 input / $5.00 output per 1M tokens); OpenAI quotes an estimated $0.003/min, used here. Deprecated: OpenAI announced this model will shut down on 2027-02-26 (https://developers.openai.com/api/docs/guides/speech-to-text); price remains valid until then. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled.
Source: developers.openai.com
GPT-4o Transcribe Diarize pricing
Speech to text. Speech-to-text with speaker labels (diarized_json output with speaker, start and end per segment). Deprecated, shuts down 2027-02-26.
| Monthly usage | Price per minute |
|---|---|
| Any volume | $0.006 |
Token-billed ($2.50 input / $10.00 output per 1M tokens); OpenAI quotes an estimated $0.006/min, used here. Includes speaker labels at no extra charge. Deprecated: OpenAI announced this model will shut down on 2027-02-26 (https://developers.openai.com/api/docs/guides/speech-to-text); price remains valid until then. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled.
Source: developers.openai.com
Whisper pricing
Speech to text. Original open-weights-derived Whisper hosted transcription model covering 98 languages. Deprecated, shuts down 2027-02-26.
| Monthly usage | Price per minute |
|---|---|
| Any volume | $0.006 |
Billed per minute; the pricing page lists "Whisper" at $0.006/min. Deprecated: OpenAI announced this model will shut down on 2027-02-26 (https://developers.openai.com/api/docs/guides/speech-to-text); price remains valid until then. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled.
Source: developers.openai.com
TTS-1 pricing
Text to speech. Lower-latency standard text-to-speech model with built-in voices. Deprecated, shuts down 2027-01-06.
| Monthly usage | Price per 1M characters |
|---|---|
| Any volume | $15.00 |
tts-1 billed per 1M input characters. Deprecated: tts-1, tts-1-hd and gpt-4o-mini-tts shut down on 2027-01-06 (https://developers.openai.com/api/docs/guides/text-to-speech); OpenAI recommends migrating to gpt-realtime-2.1-mini via the Realtime API, which is token-priced and cannot be expressed per character. gpt-4o-mini-tts (text input $0.60/1M tokens, audio output $12/1M tokens) is token-priced only and is omitted. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled. EU data residency (eu.api.openai.com, EEA + Switzerland) is available but requires Modified Abuse Monitoring or Zero Data Retention and adds a 10% uplift for models released on or after 2026-03-05 that are eligible (not included in this price).
Source: developers.openai.com
TTS-1 HD pricing
Text to speech. Higher-quality text-to-speech model with built-in voices. Deprecated, shuts down 2027-01-06.
| Monthly usage | Price per 1M characters |
|---|---|
| Any volume | $30.00 |
tts-1-hd billed per 1M input characters. Deprecated: tts-1, tts-1-hd and gpt-4o-mini-tts shut down on 2027-01-06 (https://developers.openai.com/api/docs/guides/text-to-speech); OpenAI recommends migrating to gpt-realtime-2.1-mini via the Realtime API, which is token-priced and cannot be expressed per character. gpt-4o-mini-tts (text input $0.60/1M tokens, audio output $12/1M tokens) is token-priced only and is omitted. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled. EU data residency (eu.api.openai.com, EEA + Switzerland) is available but requires Modified Abuse Monitoring or Zero Data Retention and adds a 10% uplift for models released on or after 2026-03-05 that are eligible (not included in this price).
Source: developers.openai.com
GPT-Realtime-2.1 pricing
Realtime voice. Flagship speech-to-speech realtime reasoning model with tool use and configurable reasoning effort; improved noise handling and interruption behaviour.
| Input tokens | $32.00 per 1M |
|---|---|
| Output tokens | $64.00 per 1M |
| Converted to minutes at | 600 in and 1,200 out tokens per minute |
Per-token billing on AUDIO tokens. Token conversion factors are documented by OpenAI (https://developers.openai.com/api/docs/guides/realtime-costs): audio tokens in user messages are 1 token per 100 ms (600 tokens/min); audio tokens in assistant messages are 1 token per 50 ms (1,200 tokens/min). The calculator applies both factors to every session minute, i.e. it assumes continuous speech in both directions, and ignores text/reasoning tokens, tool calls and the growing conversation context re-billed (at the cached-input rate) on every turn, so real bills can differ. Audio token rates: $32.00 input, $64.00 output, $0.40 cached input per 1M tokens. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled. EU data residency (eu.api.openai.com, EEA + Switzerland) is available but requires Modified Abuse Monitoring or Zero Data Retention and adds a 10% uplift for models released on or after 2026-03-05 that are eligible (not included in this price).
Source: developers.openai.com
GPT-Realtime-2.1 Mini pricing
Realtime voice. Faster, lower-cost realtime voice model in the gpt-realtime-2.1 family.
| Input tokens | $10.00 per 1M |
|---|---|
| Output tokens | $20.00 per 1M |
| Converted to minutes at | 600 in and 1,200 out tokens per minute |
Per-token billing on AUDIO tokens. Token conversion factors are documented by OpenAI (https://developers.openai.com/api/docs/guides/realtime-costs): audio tokens in user messages are 1 token per 100 ms (600 tokens/min); audio tokens in assistant messages are 1 token per 50 ms (1,200 tokens/min). The calculator applies both factors to every session minute, i.e. it assumes continuous speech in both directions, and ignores text/reasoning tokens, tool calls and the growing conversation context re-billed (at the cached-input rate) on every turn, so real bills can differ. Audio token rates: $10.00 input, $20.00 output, $0.30 cached input per 1M tokens. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled. EU data residency (eu.api.openai.com, EEA + Switzerland) is available but requires Modified Abuse Monitoring or Zero Data Retention and adds a 10% uplift for models released on or after 2026-03-05 that are eligible (not included in this price).
Source: developers.openai.com
GPT-Live 1 pricingNew
Realtime voice. Full-duplex voice model that listens and speaks at the same time and delegates reasoning and tool use to a backend model or agent. Voice layer only.
| Audio in | $0.05 per minute |
|---|---|
| Audio out | $0 per minute |
GPT-Live voice sessions are $0.05 per minute, billed per second with no rounding (single session rate; modelled as audio_in with audio_out 0). The backend model and tool usage that GPT-Live delegates to is billed separately at normal Responses pricing and is NOT included, so this understates the full conversation cost. Fetched 2026-10-23 from the raw HTML of https://developers.openai.com/api/docs/pricing (platform.openai.com/docs/pricing 301-redirects there; openai.com/api/pricing returns 403 to automated fetches). Free-tier credits are not modelled. EU data residency (eu.api.openai.com, EEA + Switzerland) is available but requires Modified Abuse Monitoring or Zero Data Retention and adds a 10% uplift for models released on or after 2026-03-05 that are eligible (not included in this price).
Source: developers.openai.com
Features
| Feature | OpenAI |
|---|---|
| Speaker labels | Extra cost |
| Languages | Not published |
| Self-hosting | No |
| EU data residency | Yes |
| Realtime voice | Yes |
Frequently asked questions
- How much does OpenAI speech-to-text cost per month for 1,000 minutes?
- About $4.50 a month using $0.0045/min (prices updated 2026-10-30). Free-tier credits are not subtracted.
- How much does OpenAI text-to-speech cost per month for 1,000,000 characters?
- About $15.00 a month using $15/1M chars (prices updated 2026-10-30). Free-tier credits are not subtracted.
- How much does OpenAI voice agent cost per month for 1,000 minutes?
- About $96.00 a month using Per-token billing, converted to minutes (prices updated 2026-10-30). Free-tier credits are not subtracted.
- Where does the OpenAI pricing on this page come from?
- From the provider’s own public pricing page, last checked on 2026-10-30. The methodology page lists every source and rule.