Performance and speed

Responses stream in progressively — you see the answer as it is generated instead of waiting for the whole block. Even so, some answers legitimately take longer than others, and knowing why helps you work faster.

What affects response time

• Depth of the question. Deep analysis or Research mode — which works in rounds and reviews evidence before deciding the next step — takes more time than a quick factual answer.
• Amount of material. Conversations with several documents or a large accumulated context need more processing. This is the trade-off of Lunar remembering more: richer context, slightly heavier turns.
• The study layer. Concept extraction, summaries and citation processing run around the conversation. Lunar deliberately skips those steps when they would add nothing — no concepts are extracted from a bare greeting.

[!NOTE] Lunar is designed to avoid unnecessary work: background updates only fire on real new messages, and quick local checks run before any smarter pass. That is intentional — it keeps the experience fast.

Streaming behavior

• Normal answers arrive token by token, so you can start reading immediately.
• If a response arrives all at once instead of streaming, it usually means the query was heavy — deep analysis, Research mode, or multiple documents. Split it into shorter steps and streaming normalizes.

Keeping long sessions fluid

• Ask specific questions rather than broad ones.
• Split very long topics into parts — or into microchats.
• Continue in the same thread: preserved context avoids re-explaining and produces more consistent answers than starting cold.
• Prefer Lunar Classic for quick back-and-forth, and switch to Lunar Creative when the session genuinely needs its long context (see Models).

[!TIP] If a session starts feeling slow after a very long conversation, open a branch of the current chat. You keep the relevant context without carrying every earlier turn forward.