Which AI models and model IDs does Samesoul use?

Chat models and versions

This list was checked on 2026-10-14 against the current Samesoul app's model selection and training catalogues. These are the seven named chat models; Auto is a selection mode. Check the model picker in your app for available options.

Model IDs, Auto and other identifiers

• Copy the full identifier, including its case and prefix: GLM-5.3 maps to ZHIPU/GLM-5.3 in Samesoul. Provider APIs may use different identifiers. This list does not establish a public Samesoul API or guarantee direct access to a provider API.
• Smart thinking sends the preference auto; the server chooses a model for the question and returns the selected model. Auto is a selection mode, not an eighth base model or a guarantee that every question uses Qwen.
• Qwen3-8B is a training base, not the default chat model. An ft-… value identifying a training task must not be substituted for a chat model ID. Check a trained model's deployment information separately.
• 1739309 is Samesoul's Xiaomi app store ID. It identifies the app listing, not an AI model.

How to choose a chat model

1. Open the model picker in the Samesoul chat screen.
2. Choose a named chat model, or choose Smart thinking (Auto) to let Samesoul select a model for the question.
3. Check account access and credit information, then send your message. You do not need to create a training task to switch a chat model.

Image generation

Seedream 5.0 Pro generates images from text or reference images. Select it for an image task. Its access, limits and credit costs are separate from the chat model list above.

Base models for SFT and DPO training

Checked on 2026-10-14: the training centre has these two preset base models. SFT uses example responses; DPO uses preferred and non-preferred response pairs.

• Qwen3-8B
• Qwen3-4B-Instruct-2507

• Plans and credit costs

Chat, training and memory serve different purposes

• Chat: use an available model to answer questions without preparing a training dataset.
• Training: choose a supported base model, prepare a dataset and start a separate task. Being available for chat does not mean a model can also be trained in Samesoul.
• Memory and knowledge bases: provide saved personal context or retrieve uploaded documents. They do not require you to train a base model.

• Memory and knowledge bases

Version changes and availability

• The current Qwen flagship selection is Qwen3.8-Max (qwen3.8-max). The current Zhipu selection is GLM-5.3 (ZHIPU/GLM-5.3). Use the current table above when checking versions; historical screenshots and release notes do not describe the current selection. A download page's update date alone does not prove its screenshots are current.
• Model selection can change with app updates and service availability. Check the full model name and this page's review date when comparing information.
• This page covers the Samesoul app's chat selection, image generation and preset training bases. Access, feature support and charges depend on your account and the current app; it is not a promise of unlimited use.