[Feature Request] Separate Instant, Vision, and Expert Modes While Allowing Mode Switching Within Existing Chats

The current combination of Instant, Vision, and Expert modes makes the different modes share the same chat experience. In the previous DeepSeek interface, Expert Mode was described as having “limited server resources,” and I am currently encountering those resource limitations when trying to use Expert Mode in the current system.

This makes Expert Mode less reliably available even when Instant and Vision remain usable. The three modes also have substantially different resource requirements and capabilities, so combining them does not necessarily provide the best balance between convenience and availability.

Before the change associated with Issue #1614, each chat had a selected mode. However, the main limitation of the previous system was that users could not change the mode within an existing chat.

I would like DeepSeek to revert the combination of Instant, Vision, and Expert modes and restore them as distinct modes, while allowing users to switch the mode within an existing chat.

For example, a user could start a conversation in Instant Mode, switch the same conversation to Vision Mode when visual understanding is needed, and later switch it to Expert Mode for more demanding reasoning.

This would preserve the convenience of changing modes without requiring users to start a new conversation, while keeping the modes technically separated and allowing their server resources to remain independently managed.

I have not considered simply returning to the old per-chat mode system, because that would retain its main limitation: users would still be unable to change the mode within an existing conversation.

My proposal is specifically intended to combine the advantages of the old and current systems: separate modes with the ability to switch between them within an existing chat.

Issue #1614 proposed a unified multimodal interface combining Fast/Instant, Expert, and Vision capabilities. I appreciate the convenience of being able to use different capabilities in the same conversation, but I think the modes should remain separately selectable rather than being fully combined.

The key distinction in this proposal is:

Old system: separate modes, but no mode switching within an existing chat.

Current system: combined modes, while Expert Mode has limited server resources.

Proposed system: separate modes with mode switching within an existing chat.

This would retain the main convenience of the unified approach while avoiding the current problem I am experiencing with Expert Mode's limited server resources.