GPT Realtime vs GPT Realtime 2.1

GPT Realtime is served by invitation. Its figures below are the live rates, but calls need a workspace grant first; ask us for access before you build on this comparison.

GPT Realtime 2.1 is served by invitation. Its figures below are the live rates, but calls need a workspace grant first; ask us for access before you build on this comparison.

vs

Which one, when

Text input is $4 per million on both, audio is $32 in and $64 out on both, and cache reads are $0.4 on both; gpt-realtime-2.1 charges more only on text output, $24 against $16, and quadruples the session context to 128K from 32K. It also adds configurable reasoning effort and interruption handling to the older model's function calling. Pick gpt-realtime only to hold an existing integration - on everything but text output the newer model costs the same.

Benchmarks

GPT Realtime 2.1: the vendor has not published benchmark scores.

GPT Realtime: 8 published, but no benchmark it shares with enough other models to compare.

GPT Realtime GPT Realtime 2.1 other models measured peer average ★ no peer scored higher
AMI near+far field (WER)
35.9%
N/A
Big Bench Audio
82.8%
N/A
BFCL v3 (single-turn, python only)
80.4%
N/A

Vendor-published: Amazon OpenAI

Pricing

GPT Realtime GPT Realtime 2.1 Δ
Audio input / 1M tokens $32 $32 =
Audio output / 1M tokens $64 $64 =
Audio cache read / 1M tokens $0.4 $0.4 =
Text input / 1M tokens $4 $4 =
Text output / 1M tokens $16 $24 0.67×
Cache write no separate charge no separate charge -

Rates from the live catalogue at build time; each model page carries the current rate card.

Where they sit · price per 1M audio tokens across all 7 realtime speech-to-speech models on this billing unit (log scale)

Capabilities

GPT Realtime GPT Realtime 2.1
Prompt caching implicit (automatic) implicit (automatic)
Cache lifetime 5-10m, up to 1h 5-10m, up to 1h
Minimum cached prefix varies with request settings varies with request settings

Specs

GPT Realtime GPT Realtime 2.1
Input modalities text audio text audio
Output modalities text audio text audio
Released 2025-08-28 2026-07-06
Knowledge cutoff 2023-10 2024-09
Session capabilities
  • Native speech-to-speech over WebSocket
  • function calling
  • 32K context
  • Speech-to-speech over WebSocket
  • configurable reasoning effort
  • tool use
  • interruption handling
  • 128K context
Context window 32,000 128,000

Specs are transcribed from each vendor’s documentation; a row a vendor does not publish is left out rather than inferred. Full sources: GPT Realtime · GPT Realtime 2.1

Switch between them with one line

Both ids are in every tab below; the highlighted pair of lines is the only edit. Same endpoint, same key, same request shape.

import asyncio, base64, json, websockets

URL = "wss://synthorai.io/v1/realtime?model=gpt-realtime"
# URL = "wss://synthorai.io/v1/realtime?model=gpt-realtime-2.1"  # uncomment this line, comment the one above
# Send ONLY the Authorization header - the beta protocol is retired.
HEADERS = {"Authorization": "Bearer sk-syn-..."}

async def main():
    async with websockets.connect(URL, additional_headers=HEADERS) as ws:
        # 1) configure the speech-to-speech session
        await ws.send(json.dumps({
            "type": "session.update",
            "session": {
                "type": "realtime",
                "output_modalities": ["audio"],
                "audio": {"output": {"voice": "alloy"}},
            },
        }))
        # 2) send input audio (base64 PCM16), then request a spoken reply
        await ws.send(json.dumps({"type": "input_audio_buffer.append", "audio": pcm16_b64}))
        await ws.send(json.dumps({"type": "input_audio_buffer.commit"}))
        await ws.send(json.dumps({"type": "response.create"}))
        # 3) stream the model's audio (and text) back
        async for raw in ws:
            ev = json.loads(raw)
            if ev["type"] == "response.audio.delta":
                play(base64.b64decode(ev["delta"]))   # audio out
            elif ev["type"] == "response.done":
                break

asyncio.run(main())

Get your API key →

FAQ

Which is cheaper, GPT Realtime or GPT Realtime 2.1?

Both list $32 for “Audio input / 1M tokens”. Other rates and the number of units used can still change the total cost.

Can I A/B test GPT Realtime against GPT Realtime 2.1 without two integrations?

Yes. Both are served through the same OpenAI-compatible endpoint with one API key. Switching is a one-line change to the model id, so you can route a fraction of traffic to each and compare bills directly.

Do GPT Realtime and GPT Realtime 2.1 support prompt caching?

Yes. Both bill cache reads below their input rate, so warm-prefix workloads cost less than the list rates suggest. The exact cache-read rows are in the pricing table above.

Related comparisons

From our measured studies