OpenAI Releases GPT-Live-1.1: 13 New Voices Across Six Languages, API Still Pending

On October 15, 2026, OpenAI pushed GPT-Live-1.1 to all users. The update covers three products—ChatGPT voice, Codex, and dots—adds 13 new voice presets, expands supported languages to French, German, Hindi, Mexican Spanish, Korean, and Japanese, and improves the pause experience during real-time voice search. According to the company, the API endpoint “will be supported soon,” but no specific date has been set, and developers are still using the previous model.

13 Voices, Six Languages, a Deliberate Market Choice

The 13 new voices are distributed as follows: French adds Aube and Rivage, German adds Sonna and Miko, Hindi adds Leher and Neer, Japanese adds Himari and Rindō, Korean adds Haetsal and Hanul, Mexican Spanish adds Tinta and Arrullo, and English adds Bubbie. Each new language gets two voices—not by accident. Two options mean at least some degree of personalization, rather than a language having to accept a single “official accent.”

The language choices themselves reveal a geographic strategy. Hindi points to the South Asian market; Mexican Spanish rather than Castilian Spanish means OpenAI clearly prioritizes Latin American users over the Iberian Peninsula. Japanese and Korean together cover two major high-spending technology markets in East Asia. French and German are the two most important native-language economies within the EU. Together, these six language groups represent a considerable non-English share of potential paying users worldwide.

The simultaneous update also introduces GPT-Live-1.1 Quip—a mini variant of this version designed specifically for ChatGPT voice scenarios. The existence of the mini version shows that OpenAI is addressing both performance and lightweight needs in the same voice update, consistent with its “flagship + mini” dual-track product strategy for text models.

Fewer “checking...” Prompts: The Experience Logic Behind a Technical Detail

The second core change in this update is a reduction in the mechanically repeated “checking...” prompt during voice search. GPT-Live-1.1 does not eliminate search prompts entirely; instead, it introduces a “broader phrase library” to replace the single “checking...”—in other words, the prompts have become diversified.

When real-time voice AI searches external information, network request latency is unavoidable; this is a physical limitation that cannot be eliminated. The repeated appearance of “checking...” is essentially the AI using language to fill the gap created by latency. When this filler is too monotonous and mechanical, users do not perceive “the AI is helping me find an answer” but rather “the AI is stuck”—two completely different psychological experiences. Replacing a single filler with a rich phrase library essentially uses language design to compensate for the gap in perceived experience when latency cannot be compressed.

The cost of this change is invisible, but the benefit is direct: user retention and conversation completion rates are largely affected by perceived fluency, not just actual latency in milliseconds.

What It Means to Update Three Product Lines Simultaneously

GPT-Live-1.1 is deployed simultaneously to ChatGPT, Codex, and dots, covering the main consumer interface, developer coding scenarios, and the standalone entity dots. The three products share one underlying voice model version, meaning OpenAI is pursuing a unified infrastructure strategy for voice capabilities rather than maintaining separate voice branches for different products—this reduces maintenance costs and ensures consistency, but also means that a voice defect in one product will be exposed across all three lines at once.

For enterprise users and developers, the most critical information right now is: the API endpoint remains on the old version. The official description is that “developers remain on the previous model until the API update arrives,” creating a temporary feature gap between the consumer side and the API side. Enterprises that call voice capabilities through the API cannot currently access the new voices or the improved search prompt experience; ordinary ChatGPT users, by contrast, are enjoying this update before developers. This rhythm—consumer side first, API following later—is not uncommon in OpenAI’s recent releases, but for developers who rely on the API to provide multilingual voice services in their products, it means their product upgrade cycles are constrained by an unannounced timeline.

Actionable Recommendations for Developers and Enterprise Selection

For enterprises and individual users currently using voice features directly through the ChatGPT app: this update is being rolled out seamlessly, and no action is required to enjoy the new voices and improved pause experience. If business scenarios involve customer service in the six languages above, now is the time to assess whether to incorporate voice interaction into UI design.

For developers building voice features through the API: the current recommendation is to keep existing integrations unchanged while watching for OpenAI’s API update announcements. Until the API version is confirmed to launch, it is inadvisable to promise specific delivery dates for multilingual voice features in a product roadmap, because it depends on a “coming soon” milestone with no date. During the waiting period, teams can prepare multilingual voice UX designs and voice testing workflows in advance so they can integrate quickly once the API launches.

For enterprises considering embedding voice AI in customer service or education scenarios: GPT-Live-1.1’s support for Hindi and Mexican Spanish opens new possibilities for localized services targeting Indian and Latin American markets. But until the API endpoint launches, this capability can only be implemented through the ChatGPT application layer rather than programmatic integration—a clear limitation for large-scale deployment.

Competitive Pressure in Multilingual Voice AI

The timing and language choices of GPT-Live-1.1 show that competition in voice AI is shifting from “can it speak?” to “does it sound natural enough, and is its coverage broad enough?” The diversification of voices (two options per language) is a signal: a single “official accent” is no longer enough to meet the differentiated needs of the global market, and voice personalization will become a competitive dimension in the next stage.

The delayed launch of the API endpoint is a key signal for judging the pace of OpenAI’s voice strategy. If this delay exceeds one month, it may indicate that the multilingual voice model still faces engineering challenges in API-side stability or cost control; if it follows soon, it means the consumer-side launch was a deliberate testing strategy—first tune with real user data, then open it to developers to build more complex applications. Watching this time window will be an effective indicator of whether OpenAI truly regards multilingual voice as a core API product capability.

The existence of GPT-Live-1.1 Quip suggests that tiered pricing for voice models may be in the works—lightweight versions are often a prelude to cost-differentiated pricing. If the API later launches both a flagship version and a Quip version, developers will face a new round of trade-offs among latency, quality, and cost.