x.com/ArtificialAnlys/status/1976696262083985636
1 correction found
nearly 7x cheaper than the flagship GPT Realtime
For speech-to-speech usage, OpenAI describes gpt-realtime-mini as about 70% cheaper than the large model, not nearly 7x cheaper.
Full reasoning
OpenAI’s own DevDay 2025 announcement describes gpt-realtime-mini as "70% less expensive than the large model", which is materially different from "nearly 7x cheaper."
The model pricing pages also show why. For audio tokens (the relevant modality for a speech-to-speech model), gpt-realtime-mini is priced at $10 / 1M input audio tokens and $20 / 1M output audio tokens, while gpt-realtime is $32 / 1M input audio tokens and $64 / 1M output audio tokens. That makes the mini model about 3.2× cheaper on audio, or 68.75% cheaper—close to OpenAI’s 70% figure, but nowhere near 7×.
A ~6.7× difference does exist for some text-token prices ($4 vs. $0.60 input, $16 vs. $2.40 output), but this post explicitly frames the comparison as a native speech-to-speech model, so using "nearly 7x cheaper" as the headline claim is misleading for the model’s primary audio use case.
3 sources
- OpenAI DevDay 2025 | OpenAI
### gpt-realtime-mini — A smaller voice model that’s 70% less expensive than the large model.
- gpt-realtime-mini Model | OpenAI API
Audio tokens ... Input $10.00 ... Output $20.00 ... A cost-efficient version of GPT Realtime - capable of responding to audio and text inputs in realtime over WebRTC, WebSocket, or SIP connections.
- gpt-realtime Model | OpenAI API
Audio tokens ... Input $32.00 ... Output $64.00 ... This is our first general-availability realtime model, capable of responding to audio and text inputs in realtime over WebRTC, WebSocket, or SIP connections.