Since September 9, 2026, ChatGPT Voice makes a clearer distinction between three things that were often conflated: the model running the real-time voice conversation, the model that can be called for search or harder reasoning, and the amount of Live usage included with each plan. In practice, GPT-Live-1 or GPT-Live-1 mini remain the spoken-conversation layer; GPT-5.6 or GPT-6 Astra can come into play when a task needs more reasoning or search, depending on the models available to the plan.
The change was announced in the official ChatGPT release notes on September 9. The limits below reflect the official ChatGPT Voice page as checked on September 10, 2026.
The key distinction: Live, GPT-5.6, and GPT-6 Astra are not the same layer
| Layer | What it does | Models cited by OpenAI |
|---|---|---|
| Real-time voice conversation | Listens and responds by voice, supporting continuous conversation and natural interruptions. | GPT-Live-1 or GPT-Live-1 mini, depending on the plan. |
| Search and harder reasoning | Can be used inside the conversation when a question requires search or additional reasoning. | GPT-5.6 or GPT-6 Astra, depending on the plan and available controls. |
| Usage allowance | Defines how long Live can be used within the rolling window. | Hours of GPT-Live-1 or GPT-Live-1 mini; it is not an “hour of GPT-5.6” or an “hour of Astra.” |
This distinction prevents a common misunderstanding: seeing “GPT-6 Astra” in the Voice notes and assuming Astra replaced the audio model. Current documentation says Live is still powered by GPT-Live-1 or GPT-Live-1 mini. What changed is that the conversation can draw on the text/reasoning models available to the account when needed.
New Voice limits for Go, Plus, and Pro
| Plan | Live model | Current Voice in Chat limit |
|---|---|---|
| Go | GPT-Live-1 mini | Up to 3 hours |
| Plus | GPT-Live-1 | Up to 3 hours |
| Pro — $100/month | GPT-Live-1 | Up to 15 hours |
| Pro — $200/month | GPT-Live-1 | Unlimited |
These limits are measured over a rolling 24-hour period. So “3 hours per day” does not necessarily mean the counter resets at midnight. The system uses a rolling window and notifies you when you reach the limit.
Another important change is the end of the mini fallback on paid plans: the September 9 notes say Plus and Pro no longer switch to GPT-Live mini after reaching a Voice limit. That makes the allowance easier to understand, but it also means there is no automatic second mini quota to continue the same Live experience.
What happened to Instant, Medium, and High in Voice
The separate Voice intelligence levels — Instant, Medium, and High — have been deprecated as Voice-specific controls. OpenAI now directs users to the same model and reasoning-effort controls used in text chat. In other words, the choice of how much reasoning to apply is no longer a parallel setting exclusive to Voice.
This does not mean Instant, Medium, and High disappeared from ChatGPT as a whole. They remain reasoning controls for eligible text models; what was retired was the separate version of those levels inside the Voice experience.
Which reasoning models each plan can bring into the conversation
Availability of GPT-5.6 and GPT-6 Astra follows what the account can use in ChatGPT. The official GPT-5.6 and GPT-6 Pro page helps interpret the change:
- Go: uses GPT-5.6 Luna as its default model and for Think. Go does not have access to GPT-5.6 Sol.
- Plus: includes GPT-5.6 Sol with Medium and High reasoning. Current documentation does not list GPT-6 Pro/Astra for Plus in Chat; Astra is listed for Plus in ChatGPT Work and Codex.
- Pro: includes GPT-5.6 Sol, Extra High, and Pro models; GPT-6 Pro, powered by GPT-6 Astra, is available on eligible Pro plans.
That is why the statement “Voice can use GPT-5.6 or GPT-6 Astra” should not be read as a promise of Astra on every plan. The release note itself says available models and their limits depend on the plan.
A practical example of how the new architecture works
If you are on Plus and ask a simple question by voice, the conversation still runs through GPT-Live-1. If the question requires deeper reasoning, Voice can draw on GPT-5.6 within the options allowed for Plus. On Pro, a compatible task can also gain access to GPT-6 Pro/Astra when that model is available to the account and selected or invoked through the corresponding controls.
The time counted against the Voice limit, however, remains tied to Live usage. That is why it helps to treat “which model is speaking with me?”, “which model is reasoning for this answer?”, and “how much Voice time do I have left?” as three separate questions.
How to check what applies to your account
- Open the model picker and reasoning controls used in text chat; Voice now follows those same controls for search and reasoning tasks.
- Check your plan: Go, Plus, and the two Pro price tiers have different Live allowances.
- Remember that the limit uses a rolling 24-hour window, not necessarily a midnight daily reset.
- If you use Plus or Pro, do not rely on an automatic switch to GPT-Live mini when your allowance ends; that fallback was removed.
- For information that can change, check the official Voice and ChatGPT model pages, because availability and limits can evolve.
What changed on September 9, in short
- GPT-Live-1 and GPT-Live-1 mini remain the models powering the Live experience.
- GPT-5.6 or GPT-6 Astra can be used when Voice needs search or reasoning, depending on the plan.
- Go now has up to 3 hours of GPT-Live-1 mini.
- Plus now has up to 3 hours of GPT-Live-1.
- Pro at $100/month now has up to 15 hours of GPT-Live-1; Pro at $200/month has unlimited usage.
- Plus and Pro no longer fall back to GPT-Live mini after the allowance is reached.
- The separate Instant/Medium/High Voice levels were deprecated; Voice now uses the model and reasoning controls from text chat.
