
GPT-Live-1.1 vs Sesame Preview: A Rate Card Against a Subscription
- OrcaNEWOrcaCyber: Hy4 Flash Abliterated / Uncensored2026-11-18$0.07 / $0.28 per 1M tokens · 187 tok/s
- tencentNEWTencent: Hy4 Flash2026-11-16$0.02 / $0.07 per 1M tokens · 413 tok/s
- GoogleNEWGoogle: Gemini 4 Carbon2026-11-10$2.00 / $10.00 per 1M tokens · 87 tok/s
- AnthropicNEWAnthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 per 1M tokens · 119 tok/s
- orcaNEWOrca: OrcaC2 1.02026-11-079 tok/s
- OrcaOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 per 1M tokens · 89 tok/s
- OrcaOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 per 1M tokens · 135 tok/s
- OrcaOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$2.00 / $5.00 per 1M tokens · 78 tok/s
- tencentTencent: Hy42026-11-04$0.83 / $2.50 per 1M tokens · 116 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 per 1M tokens · 114 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 per 1M tokens · 66 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 91 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 41 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447Intelligence
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 per 1M tokens · 212 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929Intelligence
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 per 1M tokens · 111 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733Intelligence
- DeepSeekDeepSeek: DeepSeek V4.1 Flash 09102026-10-17$0.15 / $0.60 per 1M tokens · 168 tok/s
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Intelligence
These two are compared constantly, and the comparison is usually a mistake. GPT-Live-1.1 and Sesame Preview both hold a conversation, both handle interruption, and both are routinely described as sounding less like a phone tree than the generation before them. But one is a metered API and the other is a subscription app, and once you notice that, most of the interesting questions about them stop being about voice quality at all.
GPT-Live-1.1 arrived on the OpenAI API on November 19, 2026, five weeks after it replaced GPT-Live-1 inside ChatGPT Voice. Sesame Preview has been shipping to consumers since May 2026 and has been on Android since early August; in October it added a price list — Free, Plus at $20 a month, Pro at $100 a month — sold through the App Store and Google Play. Neither company publishes a benchmark that lets you put the two voices on the same axis, and no independent lab has run one. Everything below is therefore about what each product is, not about which sounds better, because the second question currently has no measurable answer.
Two products, two business models
The cleanest way to see the split is to ask what you are buying.
• Access — GPT-Live-1.1 is a model ID on v1/live/sessions, callable from your own application vs Sesame Preview is an app for iOS and Android, with no public API and no model ID
• Pricing — GPT-Live-1.1 bills per second: $0.05 per active voice minute, $0.02 on standby vs Sesame sells monthly allowances: Free at $0 (30 minutes of talk time a day), Plus at $20 (2 hours a day), Pro at $100 (10 hours a day)
• Who pays — GPT-Live-1.1 is billed to a developer account, plus a separate token bill for the backend it delegates to vs Sesame is billed to a consumer through the app stores
• Voices — GPT-Live-1.1 exposes 21 named voices across English, Portuguese, German, French, Hindi, Japanese and Korean, plus prompt-to-voice creation vs Sesame offers four characters — Maya, Miles, Simone and Charlie
• Customisation — GPT-Live-1.1 lets you write session instructions, pick a delegation mode and set an idle timeout vs Sesame's agents have distinct personalities you choose between rather than a configuration surface you write against
• Open weights — GPT-Live-1.1 is closed vs Sesame's CSM-1B is Apache 2.0 on Hugging Face, and is not the model in the app

That last line is the one people get wrong. Sesame does have an open model, and it is not Preview. CSM-1B is a 1B-parameter conversational speech model released under Apache 2.0, built on a Llama-architecture backbone with Kyutai's Mimi codec, and it is a different and smaller system than the production voice in the app. Anyone who tells you they are self-hosting Sesame Preview is self-hosting something else. The gap between the demo and the download has been the central fact of Sesame's offering since the first research post in February 2025, and it has not closed.
The $0.05 minute, priced properly
GPT-Live-1.1's rate is easy to misread in both directions, so it is worth being exact. Voice sessions bill by duration at $0.05 per active minute, charged per second, with no rounding up to the minute. Since November 19, 2026, a WebRTC session can be configured to pause inference after 5 to 60 seconds without speech and bill $0.02 a minute while it waits. Backend model and tool usage is charged separately, on the token meter of whatever model you name as the delegation target. Creating a WebRTC session bills 15 seconds of voice duration at initialisation, credited back against duration once the session starts running.
Three hours of continuous open line is therefore $9.00 on the voice meter alone, before a single backend token. That is the number to hold against Sesame's Plus tier, which charges $20 a month for two hours of talk time a day — roughly 60 hours a month, or about $0.33 an hour, if you use the allowance fully. On a pure per-hour basis Sesame is dramatically cheaper, and it is cheaper for a reason that has nothing to do with generosity: you are one consumer talking to one assistant inside an app, not an application making thousands of concurrent calls against a metered endpoint.
The comparison only becomes real when you ask what you are allowed to build. Sesame's talk time is an allowance for a person. GPT-Live-1.1's minutes are a cost of goods for a product. A team evaluating the second cannot substitute the first, at any price.

What neither product gives you
Sesame's limitations are stated by the company and worth listing without softening. The app is the product; there is no API, no model ID, no rate limit, no enterprise tier and no published monetisation plan beyond the subscriptions. Sessions are capped by a daily talk-time allowance — 30 minutes on Free, two hours on Plus, ten hours on Pro — and the agents talk rather than execute: they do not book, buy or file anything on your behalf in the way an agentic API would. The app is on iOS and Android, with eyewear announced for 2027. Sesame's stated direction is research partnerships and hardware.
GPT-Live-1.1's limitations are equally concrete and run the other way. It is audio and text in, audio and text out. Images and video are not supported. It does not do the reasoning itself — it delegates to a backend model or agent you configure, through Responses delegation or client delegation — and it is reachable only through v1/live/sessions. Chat Completions, Responses, Realtime, transcription, translation and speech generation are all listed as unsupported endpoints for this model. If you want a single call that returns text, this is not it; it is a voice layer with a hole in the middle where your backend goes.
Both products, in other words, are narrower than their marketing reads. Sesame sells a voice you cannot build on. OpenAI sells a voice that cannot think on its own.
Where the numbers stop
Neither voice has an independent score, and it is worth being blunt about how complete that gap is. Artificial Analysis' speech-to-speech index, read on November 20, 2026, carries GPT-Live-1 twice — 83.2 with an Astra-class backend at medium reasoning effort and 80.1 with a Sol-class backend at low — and does not carry GPT-Live-1.1, which is a day old on the API. It carries no Sesame model at all. There is no third-party measurement of either current voice, so any claim that one sounds more natural than the other is a listening impression presented as a finding.
What does exist for Sesame is a benchmark it built. In August 2026 the company released TurnBench, an open benchmark for turn-taking in spoken dialogue: 30 hours of studio-recorded, dual-channel English conversation across 154 dialogues and 106 voice actors, triple-annotated, measuring how accurately and how quickly a system detects end-of-turn and interruption events. It is a genuine contribution — the field had been leaning on Switchboard recordings from the early 1990s — and the public leaderboard ranks 27 systems by recall with false-positive rate and median latency alongside. Neither GPT-Live-1.1 nor any Sesame production voice appears on it. The benchmark measures the problem both products claim to solve, and neither has submitted.
That is the state of the evidence: a vendor-run human evaluation on one side with no figure attached, a day-old API model on the other with no independent score, and an open leaderboard that could compare them sitting empty of both.
What OrcaRouter does and does not cover
OrcaRouter does not serve either product. There is no full-duplex voice endpoint in the catalogue, so if you want GPT-Live-1.1 you call OpenAI's Live sessions API directly, and if you want Sesame Preview you download an app. Saying otherwise would be a lie about what we host.
What we do cover is the part of a Live deployment that is a normal model call. GPT-Live-1.1 delegates reasoning and tool use to a backend, and OpenAI's own guidance is to start with gpt-6-luna and move to gpt-6-sol for harder work. Both are on OrcaRouter at provider list price with no markup, alongside the rest of the GPT-6 family and more than 200 other models behind one key, with automatic failover across providers and a routing DSL for composing several models into a single call. If your voice agent's brain is the part you want to route, switch or fail over, that is the layer we can actually help with — and the same is true of the text-to-speech side, where gpt-4o-mini-tts and Google's Gemini TTS models are also available.
The decision, stated plainly
If you are building a product, the choice is not close and it is not about voice quality. GPT-Live-1.1 is callable, priced per second, and configurable; Sesame Preview is a consumer app with a monthly allowance and no way in. Ship on the first, and treat the second as evidence of where conversational voice is heading rather than as a component.
If you are a person choosing an assistant, the comparison inverts. Sesame's $20 tier buys roughly 60 hours of talk time a month, which is about the price of six hours forty minutes on GPT-Live-1.1's voice meter, and it comes with characters designed to be talked to rather than a session configuration designed to be written. The API is not the better product for that job; it is simply the only one that lets you build.

What would change the answer is a public Sesame endpoint or an independent measurement of either voice. Neither exists as of November 20, 2026. Until one does, the useful thing to say about GPT-Live-1.1 versus Sesame Preview is that they are not really competing — and that the fact they get compared at all says more about how young this category is than about either company's work.
