GPT-Realtime-2.1

openai/gpt-realtime-2.1

GPT-Realtime-2.1 on Sylphx Models · 128K of context. One Responses contract, with list prices published per million tokens.

Released: Jul 6, 2026 · 80 days old
Provider: OpenAI
Context: 128,000 tokens

Open in the playground See a request sample Read the docs

Input $4.40 / MOutput $26.40 / MCached input $0.44 / MFull rate sheet

Context window128,000tokens per request

Max output32,000tokens per response

Input$4.40 / Mper million tokens

Output$26.40 / Mper million tokens

Cached input$0.44 / M90% off input

ReleasedJul 6, 202680 days old

Pricing

USD per million tokens

Input: $4.40
Output: $26.40
Cached input: $0.44

$9.90 for 750,000 input and 250,000 output tokens.

Assumption: one batch of 750,000 input plus 250,000 output tokens at the published list prices — recompute for your own mix. Blended $9.90 per million at a 3:1 mix.

Chat replies · 3:1 input:output: $9.90 / M
Long documents · 9:1 input:output: $6.60 / M
Heavy output · 1:1 input:output: $15.40 / M
Cache write: Not published
Currency and unit: USD / million tokens

Cached input reads cost 90% less than fresh input tokens.

Full rate sheet and billing rules.

Specification

Context window: 128,000 tokens
Max output: 32,000 tokens
Input modalities: Not published
Output modalities: Not published
Family: GPT Realtime
Version: v2.1
Provider: OpenAI
Released: Jul 6, 2026 · 80 days old
Model id: openai/gpt-realtime-2.1

Data posture

Standard

By default, prompts and outputs sent to this model are not used to train models.

Training on what you send · default: No
Zero Data Retention · default: Yes

POST https://api.sylphx.ai/v1/responses

curl https://api.sylphx.ai/v1/responses \
  -H "Authorization: Bearer $SYLPHX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-realtime-2.1",
    "input": "Summarise this incident report in three bullets.",
    "stream": false
  }'

How to call it

Send the official Responses document to the base URL above with your organization key and this exact model id. Streaming arrives as ordered events with one terminal event.

Retries: send an Idempotency-Key and a retry replays the original response instead of billing twice.
Errors: typed envelopes tell you whether to retry, wait, or fix the request.
Keys: mint and revoke organization keys from the console.

Quickstart Responses API Errors and retries

Related models

Derived from this catalog: same provider first, then the same family, then the nearest neighbours by input price and context window.

More from OpenAI

Same provider

Similar price and context

Nearest by input price and context window

Independent evaluation

We do not publish benchmark scores for this model — we have not verified any. These external leaderboards run their own tests, so check them against your own workload.

Not our numbers

Every OpenAI model →