GPT-Live-1.1: New Voices, Accents, and a Debate Over What Happened to Sol

⚡ The gist of the debate

OpenAI has rolled out GPT-Live-1.1 — an update for ChatGPT, Codex, and dots: 13 new voices, noticeably better accents in French, German, Hindi, Mexican Spanish, Korean, and Japanese, plus a promise to show "checking…" less often. The thread on OpenAI Community was started by VeitB; it has 43 replies, 119 likes, and 6,122 views. The debate isn't about whether new voices are needed, but about two things: when all of this will reach the API, and why some users' familiar voices now sound worse.

🗣 What participants are saying

Thread author VeitB writes that the update is rolling out to all users today, while the API is "coming soon." He also breaks down delegation: native integration with the Decisions API allows decisions to be made with low latency while the user is still speaking, and "native delegation skips a round trip through your app." Custom voices, he says, are built from a text description or from an approved recording; later he added a link to a video on voice setup to the top post.

The most practical question came from fred5: "it's seems that it's not yet available in the API?" VeitB confirmed and replied: "Yes, it's coming soon to the API. Fingers crossed that's very soon!" For anyone building voice agents, this means the new voices and delegation remain a showcase for now, not a working tool.

merefield praises exactly what was promised: "The current models seem to bleed one accent into another, this is great news" — and adds that humans, after all, do the same thing. On "checking…" he briefly quips: "Praise be!" And he immediately steers the conversation back to business — waiting for the API rollout.

The first notable downside comes from frz.lunarwolf: in standard mode, the Arbor voice doesn't work for live voice for him, only for read aloud. And he doesn't like the new versions of Arbor in Advanced and Live — "sound terrible to me." That's a personal assessment, with no recordings or measurements, but it shows that "13 new voices" isn't necessarily a plus for every specific preset.

nathanjay goes further and asks what happened to the Sol voice in standard mode. His second complaint is behavior: the model has started describing the aspect ratio, maximum width, and other settings of displayed images and charts. He can't attach audio, video, or an external link, so he points to his reply to an OpenAI tweet — search for the phrase "WHAT HAVE YOU DONE TO SOL?!!."

The only comment off to the side of the issues is from aprendendo.next: "This looks fun! The Marvin robot voice is a classic!" Short, but it's part of the picture too: some see the update as entertainment rather than a regression.

⚖️ Where they agree, where they diverge

The common denominator is narrow but real: there are more voices, "checking…" is promised to appear less often, and the accents in non-English languages are what people were waiting for. Everyone writing here about development wants the same thing — the API. On that point the thread isn't arguing, it's waiting.

The real point of conflict isn't the feature list, but the quality of specific voices and the model's behavior. The announcement talks about accents and controllability in general, while the complaints from frz.lunarwolf and nathanjay are about Arbor and Sol in specific modes. There's no agreement here: one participant is happy, others hear a degradation. No measurements or blind before/after comparisons are visible in the thread — except for the video nathanjay links to, which can't be attached here.

🧪 What to check yourself

• Check whether GPT-Live-1.1 is available in your account and whether it has appeared in the API: in the thread at the time of the discussion, it was only "coming soon."
• Run the same text in French, German, Hindi, and Japanese and listen to whether one accent bleeds into another.
• Compare Arbor and Sol in Standard, Advanced, Live, and read aloud modes — that's exactly where participants see the difference.
• Count how often "checking…" pops up and how it affects latency in live dialogue.
• Check whether responses have started describing the aspect ratio and width of images and charts — that's what nathanjay pointed out.

🔗 Sources

OpenAI Community — GPT-Live-1.1 is rolling out with new voices and better accents

Disclaimer: this is a summary of a discussion, not facts verified by the editorial team. Opinions belong to the participants; the numbers belong to those who cited them.