Prediction markets just beat the press again.
Polymarket posted on X that OpenAI is unveiling “The All-New ChatGPT Voice” at 10 AM PT today. The post went up at 8:19 AM PT, a full hour and forty minutes before the scheduled announcement.
This isn’t a rumor from an anonymous leaker. Polymarket is a regulated prediction market with a track record of accurate calls. When its markets resolve, the platform posts the result as breaking news. And the market that just resolved is almost certainly the GPT-Bidi-1 question.
Here is what that means for anyone who uses ChatGPT Voice.
What Polymarket said
Polymarket’s post was short and direct. It said OpenAI is unveiling “The All-New ChatGPT Voice” at 10 AM PT.
The post was timestamped July 8, 2026, 8:19 AM PT. Polymarket’s account is verified. The platform has 1.7 million followers. When a prediction market with a seven-figure volume in OpenAI-related contracts posts a scoop, it deserves a different level of attention than your average leak account.
What is the “All-New ChatGPT Voice”?
The product name Polymarket used is “The All-New ChatGPT Voice.” That’s almost certainly the consumer-facing name for GPT-Bidi-1, the bidirectional voice model that has been leaked and analyzed since mid-June 2026.
GPT-Bidi-1, internally codenamed “Maple,” is OpenAI’s full-duplex voice model. Unlike the current ChatGPT voice mode, which works like a walkie-talkie (you talk, the AI listens, then it talks while you wait), Bidi-1 operates like a phone call. Both people can speak at the same time. It handles interruptions naturally. It can pivot mid-sentence.
Based on multiple independent leak sources, the model ships with three intelligence tiers: Instant, Medium, and High. The Instant tier gives you fast, low-latency responses for quick back-and-forth. The High tier brings more reasoning depth at the cost of response time. Leaked audio samples from June 24 showed the model handling interruptions gracefully, stopping mid-countdown at 7 and pivoting to a new question within roughly 200 milliseconds.
The TestingCatalog report from June 16, 2026 first spotted the GPT-Bidi-1 identifier in code references. By June 23, UI strings for translation features surfaced. By July 2026, the model was being flagged across multiple outlets as OpenAI’s biggest voice upgrade in months.
Why Polymarket had the story first
Polymarket ran a market on this exact question. The contract was “Will GPT-Bidi-1 be released by July 31, 2026?” Traders piled in as leak evidence mounted through late June. When the market resolved in favor of a near-term announcement, the platform pushed the result as a breaking news post.
This is the same pattern Polymarket has used before. When OpenAI launched GPT-5.5 in April 2026, the Polymarket contract hit 100% YES before the official announcement went out. The platform has become a de facto news wire for AI events, especially when insider knowledge flows through prediction market trading activity.
What this means for ChatGPT users
If GPT-Bidi-1 launches as “The All-New ChatGPT Voice,” it changes how natural ChatGPT conversations feel in a real way.
Current ChatGPT Voice mode works fine for simple commands and quick questions. But it breaks down in anything approaching a real conversation. You can’t interrupt naturally. You can’t speak and be heard while the AI is responding. There’s always a gap between turns that makes the interaction feel mechanical.
Full-duplex fixes that. You can interject, the model adjusts on the fly, and the back and forth starts to resemble an actual human conversation rather than a voice-controlled search engine.
The three intelligence tiers also give you flexibility. Need a fast answer with low reasoning overhead? Instant tier. Need the model to work through a complex problem out loud? High tier. The structure mirrors the existing ChatGPT text tiers, and it lets people trade speed for depth on a per-conversation basis.
TestingCatalog also noted that ChatGPT users would likely keep the existing Advanced Voice Mode alongside the new Bidi mode, switching between them as needed rather than being forced into one or the other.
What remains unclear
Several things are still unclear.
The name may change. Polymarket called it “The All-New ChatGPT Voice,” but OpenAI hasn’t confirmed the branding. The model identifier is GPT-Bidi-1, but the consumer product name could land differently.
Pricing is unknown. Current ChatGPT Voice is included in existing subscription tiers. OpenAI may keep that model, or it could reserve the “All-New” version for Plus and higher. The leaked three-tier structure suggests some kind of usage-based differentiation, but nothing is confirmed.
Availability at launch is also a question. Some new AI features roll out to all users immediately. Others start with Plus or Team and expand over weeks. The 10 AM PT time suggests a product announcement with immediate or near-immediate availability, but I haven’t seen confirmation either way.
And of course, the full-duplex audio quality is unverified until real users test it. Leaked demos look impressive. But leaked demos have burned people before.
The bigger picture
This announcement lands at an interesting moment in the voice AI race.
xAI launched the Grok Voice Agent Builder a week ago, a no-code platform for production voice agents at $0.05 per minute with telephony tools and guardrails. Anthropic has been talking about voice capabilities for Claude but hasn’t shipped a native voice mode at ChatGPT’s scale. Google remains a player through Gemini voice, but its consumer adoption lags behind ChatGPT by a wide margin.
OpenAI has been the clear leader in mainstream voice AI usage since Advanced Voice Mode launched. What it has not done is push the underlying voice model forward at the same pace as its text models. GPT-Bidi-1 closes that gap. A bidirectional voice model with interrupt handling and multi-tier intelligence is not a minor version update. It is a structural change in what a voice AI assistant can feel like in practice.
Bottom line
Polymarket just scooped OpenAI’s own announcement by over an hour. That alone says something about how prediction markets have become a primary news source for AI industry events.
The “All-New ChatGPT Voice” announcement is set for 10 AM PT today. I will update this post with OpenAI’s official details once they drop. In the meantime, everything points to GPT-Bidi-1, full-duplex voice, interruption support, and three tiers of intelligence depth.
This is the biggest ChatGPT Voice update since the feature launched. If the demos hold up, it changes the conversation.
Update: OpenAI just confirmed it
While this article was being written, OpenAI posted its official announcement: “The next generation of ChatGPT Voice is here.” The livestream starts at 10 AM PT. Below is the official post.
That confirms everything Polymarket signaled earlier today. The “next generation” wording matches the GPT-Bidi-1 / bidirectional voice model that has been leaking for weeks. I will have full coverage and hands-on impressions once the livestream wraps.
