GPT-Bidi-1 is OpenAI’s bidirectional voice model for ChatGPT — the first ChatGPT voice model that can listen and speak simultaneously, handling mid-conversation interruptions without freezing or restarting. It launched alongside GPT-5.6 on June 26, 2026, and is available to ChatGPT Pro and Plus subscribers. Real-time translation and three speed/quality tiers (High, Medium, Instant) are included at no extra cost above the subscription price.
What GPT-Bidi-1 Does — The Three Capabilities That Change Everything
Previous ChatGPT Voice Mode worked one direction at a time: the model spoke, you waited, then you responded. GPT-Bidi-1 breaks that pattern across three specific capabilities.
Bidirectional listening. The system listens and speaks at the same time. When you interrupt mid-sentence, it stops, acknowledges the change, and continues with your correction incorporated — no freeze, no restart, no awkward pause. Testing Catalog’s launch-day demo showed a user asking Bidi1 to count to ten; mid-count, the user interrupted with a request to reverse. The model replied “okay” and immediately switched to counting down — behaviour that does not exist in any current competing AI voice product.
Real-time voice translation. Speak in English, and the model can respond in your target language as you talk. This is processed server-side rather than as a post-hoc translation layer, which reduces latency noticeably compared to piping voice output through a separate translation API.
Three intelligence tiers. High, Medium, and Instant tiers let users trade reasoning depth for lower latency. Instant is optimised for sub-300ms response starts — fast enough for natural turn-taking in conversation. High tier applies more compute and is suited to complex explanations or multi-step voice queries. Medium is the default for most interactions.
How to Access GPT-Bidi-1
GPT-Bidi-1 is available inside the ChatGPT mobile app (iOS and Android) and via the ChatGPT web interface at chatgpt.com. ChatGPT Pro subscribers ($200/month) received access at launch on June 26, 2026. ChatGPT Plus subscribers ($20/month) received it within the following week. Free-tier users have not received access as of July 2026, though the Instant tier’s low-compute design suggests a rate-limited free-tier path may be planned.
For developers: OpenAI’s platform has opened GPT-Bidi-1 via the Realtime API, consistent with how OpenAI has handled previous model rollouts. API pricing follows the existing Realtime API structure (per-second audio in/out), not a flat subscription rate.
GPT-Bidi-1 vs. Previous ChatGPT Voice Mode
The most significant change is interruption handling. Previous Advanced Voice Mode required a full end-of-turn detection before accepting new input — under load, this produced 800ms–1.5s dead zones after an interruption. GPT-Bidi-1 eliminates that dead zone. Translation was not available in previous Voice Mode without third-party tools. The three-tier quality system is also new: Advanced Voice Mode had one quality level with no user-controlled trade-off.
Why This Matters — The AI Voice Race in 2026
Voice is the competitive frontier. Google Gemini Live handles natural conversation with interruption support. Apple’s Siri in iOS 27 promises tighter on-device integration. OpenAI’s response is Bidi1: not just a model that responds to voice, but one that genuinely listens while it speaks. Since launch, ChatGPT voice has pulled ahead of most current alternatives — including Claude, which has no equivalent real-time voice product.
The broader context: the battle between ChatGPT, Claude, and Gemini in 2026 is increasingly fought on product quality rather than model benchmarks. A voice mode that behaves like a human listener is a retention driver, not just a feature — and it justifies the $20/month Plus subscription for a wider set of users than any text-only capability.
Our Take
Bidi1 fixes the biggest frustration with AI voice — the robotic wait-and-respond rhythm that makes it feel like a phone tree. Natural interruption handling and real-time translation are not incremental features; they are the floor for a voice assistant anyone would actually use hands-free. Now that it has shipped and been in real use for weeks, early user reports confirm the low-latency Instant tier holds up under real conditions. This is the most meaningful ChatGPT voice upgrade since Advanced Voice Mode launched in 2024, and more capable than what OpenAI was shipping in May.
Frequently Asked Questions
What is GPT-Bidi-1?
GPT-Bidi-1 is OpenAI’s bidirectional voice model for ChatGPT. Unlike the previous Advanced Voice Mode, it listens and speaks simultaneously, allowing natural mid-conversation interruptions without freezing or restarting. It launched on June 26, 2026 alongside GPT-5.6.
When did the new ChatGPT voice mode launch?
GPT-Bidi-1 launched on June 26, 2026. It is available to ChatGPT Pro ($200/month) and Plus ($20/month) subscribers. Free-tier access has not been announced as of July 2026.
Is GPT-Bidi-1 available via API?
Yes. GPT-Bidi-1 is available to developers via OpenAI’s Realtime API, billed per-second of audio input and output. It is not exclusive to the ChatGPT consumer interface.
What languages does GPT-Bidi-1 translate?
OpenAI has not published a full list of supported translation language pairs. The June 2026 launch demos showed English-to-Spanish and English-to-French translation in real time. Expect the supported set to expand as the model rolls out more broadly.
How is GPT-Bidi-1 different from Google Gemini Live?
Both support real-time conversation with interruption handling. The key differences are ecosystem (ChatGPT vs. Google apps), translation coverage, and the three-tier quality system that Bidi1 offers. Gemini Live is built into Android and Google Workspace; Bidi1 operates inside the ChatGPT app and API.
