ChatGPT Realtime in Russia: Why One Developer Chose Yandex

On July 6, 2026, OpenAI released ChatGPT Realtime, a speech-to-speech engine with support for interruptions and natural intonation.

What did the latency measurements show?

The author measured end-to-end latency for all three options using the same method and got a clear picture.

• Yandex Realtime — ~330 ms (direct connection in Russia, no VPN)
• OpenAI gpt-realtime — ~740 ms (only via VPN; not available directly from Russia)
• Old three-stage pipeline (STT → LLM → TTS) — ~4.4 seconds

Yandex beats OpenAI by a factor of two purely because of geography: a VPN adds several hundred milliseconds that no code optimization can remove. The three-stage pipeline loses to both by a wide margin. A 13x gap against Yandex is far too large for a voice interface, where a pause longer than 500 ms feels like the bot is "lagging."

Where both engines stumble

With OpenAI, the developer immediately hit a 400 endpoint error. It occurs when the WebSocket request is structured incorrectly and is barely documented anywhere. Debugging took a significant part of the evening.

The second problem is more serious: the engine handles Russian accents poorly. The language model was trained mostly on English, and no system prompt tuning can fix that. The limitation is architectural.

With Yandex, the barrier to entry is higher at the authorization level: three access roles and a closed OAuth→IAM exchange that is not described in detail anywhere. The documentation assumes the developer already works inside the Yandex Cloud ecosystem. Without understanding this scheme, getting a first working token is not trivial.

"Yandex handles Russian natively.

OpenAI does not, and that is a fundamental problem for a voice product on the Russian market," the author sums up.

How the migration to Yandex Realtime went

After the measurements, the decision was obvious: the author rewrote production in three key areas.

WebSocket bridge instead of REST. Yandex Realtime works over a persistent WebSocket connection rather than one-off HTTP requests. This required rewriting the audio-sending logic: live PCM streaming instead of chunked upload.

Live PCM player instead of MP3. The previous setup encoded responses as MP3 before playback, which added latency. Working directly with PCM removed that step and eliminated artifacts during interruptions, because the MP3 codec could not "close" the frame in time on a sudden stop.

Interruptions without headphones. Yandex Realtime supports echo cancellation at the engine level, so the bot reacts correctly to interruptions even through ordinary speakers. In the old three-stage pipeline, this was fundamentally impossible.

Authorization turned out to be the most labor-intensive part: the author had to work out the OAuth→IAM token exchange by hand, because there are almost no ready-made Python examples.

What this means

For voice products on the Russian market, the choice between OpenAI Realtime and Yandex Realtime is above all a choice between the convenience of a first prototype and the quality of Russian in production. Yandex wins on latency and pronunciation but requires figuring out IAM. The old three-stage pipeline is on its way out: 4.4 seconds of waiting is no longer a conversation but an answering machine from 2005.

Frequently asked questions

Can you use OpenAI Realtime in Russia without a VPN?

No. As of July 2026, the service is not available directly from Russian IP addresses. A VPN inevitably adds latency, raising it from a potential ~400 ms to ~740 ms in real-world conditions.

How does Yandex Realtime differ from regular SpeechKit?

Yandex Realtime is a full-duplex speech-to-speech engine on a single persistent WebSocket connection with support for interruptions. SpeechKit consists of separate STT and TTS services that you have to stitch together yourself into a three-stage pipeline with 4+ seconds of latency.

Is ChatGPT Realtime available in Russia?

Only via VPN, with a latency of ~740 ms; a direct connection from Russia is not available.

Why did the developer choose Yandex over ChatGPT Realtime?

Yandex Realtime showed a latency of ~330 ms from Russia, almost twice as fast as OpenAI gpt-realtime over a VPN (~740 ms).