Model Versions

Each chat model carries a version. This page is the history of every version, newest first: what each one added, and what to check before you move to it.

All versions

This page lists what only a given version and the ones after it have. Improvements that reach every version at once, such as new reply languages, are announced in the changelog.

Version 2.2

chat-2.2-smart, chat-2.2-pro and chat-2.2-auto. Released 10 September 2026. Latest version.

New

• A question control that works: the new replyParameters.questionFrequencyRate replaces the questionFrequency number with five named steps, NONE, LOW, MEDIUM, HIGH and FULL, and each step is honoured. On 2.1 the difference between the two extremes was not observable; on 2.2 it is.

Before you switch

• questionFrequency still works and still defaults to 0.75, so nothing breaks. Send questionFrequencyRate when you want the named steps; a request carrying both follows the rate.
• If you keep sending questionFrequency, it is now followed firmly: 0.5 means she asks on about one reply in two and asks nothing on the others, where earlier versions only nudged her. Sending questionFrequency: null makes her ask on every regular reply.

Version 2.1

chat-2.1-smart, chat-2.1-pro and chat-2.1-auto. Released 6 August 2026.

New

• Photo and video sharing modes: NORMAL and EAGER now differ on explicit content.
• First reply follows the conversation: she answers in the tone the User set, instead of opening with small talk.

Before you switch

• On EAGER nothing changes. On NORMAL, if you rely on nude content being shared, switch to EAGER first.
• A flirty or sexual opener now gets answered in kind, as far as sexualDrive allows.

Version 2.0

chat-2-smart, chat-2-pro and chat-2-auto. Released 12 December 2025. Version 2.0 has no .0 in its model id.

New

• SoulEngine
• Core Guardian
• Vision Link
• Voice Messages
• Smart Share
• Meeting Policy
• Engage
• Fictional Disclosure
• Platform Information
• Memory Webhook
• User Suggestions
• Multi-message replies
• All supported languages

Chat 2 model names

Where a guide says a feature works on every Chat 2 model, it means every name below.

Switching version

Set replyParameters.chatModel to chat-2.2-auto, chat-2.2-smart or chat-2.2-pro. Nothing to migrate: the next request runs on the version you name. Coming from 2.0? Read the Before you switch notes of 2.1 and 2.2: moving several versions at once brings every change in between.

chat-latest-smart and chat-latest-pro resolve to the newest version released on this page, so you pick up the next one with no change on your side — today they resolve to 2.2. There is no chat-latest-auto: an auto model always names its version, so moving it to the next one is a change on your side.

The Chat 1 models are retired: a request naming dating-smart-1 runs on chat-2-smart, and dating-pro-1 on chat-2-pro, both billed as 2.0.

Every release is also announced in the changelog.

Smart, Pro and Auto

Every version comes as three models: -smart, -pro and -auto. Use -auto: it runs each conversation on Smart for its first 30 messages, then on Pro for the rest of it.

That is exactly where Pro starts to matter. Up to 30 messages, the Persona still sees the whole conversation in her short-term memory, so Smart is enough. From the 31st, the oldest messages start leaving that window, and Pro takes over. On that first Pro reply, Auto also analyses the opening exchanges, so facts said before the switch can be memorised. Stay on Smart and nothing is ever memorised. See Conversation Memory for the details.

The User notices nothing, and each reply is billed at the tier that produced it. See Choose Your Model for the details, and for when plain Smart or Pro is the better choice.