OpenAI's upcoming 'bidi' voice mode was shown in a leaked demo posted by X user @kimmonismus on February 14, 2026. The demo demonstrates real-time interruption handling, a feature that sets it apart from current voice assistants.
Key facts
- Demo posted by @kimmonismus on Feb 14, 2026.
- Latency under 320 milliseconds per source.
- Handles interruptions and topic shifts in real-time.
- No official release date or pricing announced.
- Based on internal build, not public product.
The demo, posted as a video on X, shows a user interrupting the AI mid-sentence to ask a completely different question. The system switches context instantly, maintaining a natural conversational flow without noticeable delay. According to @kimmonismus, the latency appears to be under 320 milliseconds, comparable to human reaction time.
This 'bidi' mode — short for bidirectional, per the source — represents a significant upgrade over OpenAI's existing Advanced Voice Mode, which handles interruptions poorly and often requires the user to wait for the AI to finish speaking before responding. The demo suggests the new system can parse overlapping speech and prioritize the user's new intent, a technical challenge that has plagued voice assistants since Siri's debut in 2011.
OpenAI has not officially confirmed the name 'bidi' or provided any release date or pricing. The company's website shows no mention of the feature, and the demo appears to be from an internal build. This follows a pattern of OpenAI leaking upcoming features through social media — similar to the December 2025 leak of GPT-4.5's multimodal capabilities.
Why It Matters
The ability to handle real-time interruptions is a key differentiator for voice AI. Current systems from Google, Amazon, and Apple require explicit wake words or pauses to process new inputs. If OpenAI delivers on this promise, it could make voice interactions feel genuinely human-like, opening up use cases in customer service, live translation, and hands-free computing.
However, the demo is unverified and may not represent the final product. OpenAI has a history of demoing impressive features that take months to ship — GPT-4's vision capabilities were shown in March 2023 but didn't launch until September 2023.
Key Takeaways
- Leaked demo shows OpenAI 'bidi' voice mode handling interruptions with sub-320ms latency.
- No official release date or pricing announced.
What to watch
Watch for an official OpenAI blog post or developer event in Q1 2026. If 'bidi' mode ships, the key metric is latency under 200ms at scale and whether it supports custom voices or only preset ones. Also track Google's response — Gemini Voice Chat is their closest competitor.
