GPT-Realtime-2.1 API reference

GPT-Realtime-2.1 reasons over spoken requests, supports function tools and responds with streaming speech or text.

Call GPT-Realtime-2.1 from your own code with one PopNow API key. Below are the exact model IDs to send, every input field the model accepts, what each request costs in USD, and a request you can paste into a terminal.

At a glance

Made by: OpenAI
Output: Audio
From: $0.0048 USD per_1k_tokens
Model IDs: openai/gpt-realtime-2.1/realtime

Prices are in USD, charged to your separate API wallet. A task reserves the maximum possible charge when accepted and settles actual usage on completion; any remainder is released. Failed tasks are not charged. Call POST /api/v1/jobs/quote for an exact quote before running.

Realtime Voice

openai/gpt-realtime-2.1/realtime

Price

Option	USD
`Default`	$0.0048 per_1k_tokens

Parameters

Field	Type	Default	Allowed	What it does
`prompt`Optional	`string`	—	up to 4,096 characters	Text that starts a single spoken response; live audio uses the Realtime WebSocket.
`voice_id`Optional	`string`	—	alloy, ash, ballad, coral, echo, sage, shimmer, verse, marin, cedar	Preset speaking voice; select before the first audio response.
`reasoning_effort`Optional	`string`	—	minimal, low, medium, high, xhigh	Reasoning effort for the spoken response; reasoning consumes output tokens.
`output_format`Optional	`string`	—	pcm, g711_ulaw, g711_alaw	Live playback encoding; single-response tasks return a matching WAV container.
`speed`Optional	`number`	`1`	0.25 to 1.5	Speaking-speed multiplier for output audio.
`max_tokens`Optional	`integer`	`512`	16 to 32000	Per-turn output cap, including reasoning, text and audio tokens.
`max_context_tokens`Optional	`integer`	`4096`	512 to 123888	Retained-history cap; this, the 4096-token instruction allowance and output must fit 128000.
`audio_input_format`Optional	`string`	`pcm`	pcm, g711_ulaw, g711_alaw	Encoding of live microphone samples; PCM is 24 kHz mono 16-bit, G.711 is 8 kHz.
`audio_noise_reduction`Optional	`string`	`near_field`	near_field, far_field, none	Input denoising profile for a close microphone, room microphone, or preprocessed audio.
`realtime_session`Optional	`object`	—	—	Advanced native session configuration; overlapping regular controls must agree. Function tools are executed by your application.

Example request

This request runs as it is; any field you leave out uses its default. Put your key in POPNOW_API_KEY first.

POST /api/v1/jobs/createTask

curl -X POST "https://api.popnow.ai/api/v1/jobs/createTask" \
  -H "Authorization: Bearer $POPNOW_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
  "model": "openai/gpt-realtime-2.1/realtime",
  "input": {
    "prompt": "A bright, cheerful scene with soft natural light"
  }
}'

The response carries a taskId straight away. Read the result with GET /api/v1/jobs/recordInfo, or add callBackUrl to be notified.

GET /api/v1/jobs/recordInfo

curl "https://api.popnow.ai/api/v1/jobs/recordInfo?taskId=TASK_ID_FROM_CREATE_TASK" \
  -H "Authorization: Bearer $POPNOW_API_KEY"