Roleplay
Users praise the dialogue and prose, while others report overthinking, instruction drift, or a scene that gets steered back toward safer ground. Test your own card.
English
KIMI K3 + SILLYTAVERN
Kimi K3 SillyTavern
Looking for Kimi K3 in SillyTavern usually means three questions: does the prose hold a character, how do I connect the API, and why is it thinking or repeating so much? Here is the short answer, then a route to try Kimi K3 in Tabbit without wiring an endpoint first.
Kimi K3 is the official model name. Moonshot’s API alias is kimi-k3. The API launch was July 16, 2026.
FIRST, THE REALITY CHECK
K3 can write well and still cost patience
The official docs and community reports point in the same direction: K3 is a large, always-thinking model. RP quality depends on the prompt stack and provider as much as the model label.
Thinking and speed
Thinking is always on in Moonshot’s API. Public X posts describe slow replies, 429s, and small quotas. Those are user reports, not an uptime promise.
The model facts
Kimi K3 is a 2.8T-parameter model with a 1M-token context window. The API supports low, high, and max reasoning effort; the exact provider limits vary.
SILLYTAVERN PATH
A clean Kimi K3 setup order
Keep each layer separate. When a reply looks wrong, you can tell whether the endpoint, template, preset, or thinking history is responsible.
Preset names and extensions come from community work. Download current files from their authors; this page does not republish them.
- 01
1. Pick a provider that lists K3
Use Moonshot’s API or a provider that explicitly exposes the model. Check the current model list, base URL, billing, rate limits, and model slug before touching SillyTavern.
- 02
2. Select Chat Completions
K3’s official quickstart uses OpenAI-compatible Chat Completions. Do not paste a text-completion template into the connection and assume the model is broken.
- 03
3. Start with short instructions
Community answers repeatedly recommend clear, compact instructions. Add character, scene, style, and boundary rules to the system prompt instead of stacking contradictory patches.
- 04
4. Treat prefill as an extension, not a cure
Kimi Thinking Prefill and preserved-thinking discussions are clues for troubleshooting reasoning blocks and turn-to-turn context. Use the current extension documentation, then test one change at a time.
A LOWER-FRICTION ROUTE
Use the model beside the page in Tabbit
If your immediate goal is “let me talk to Kimi K3 while reading a character sheet or wiki”, Tabbit removes the API endpoint step. Its current model UI includes Kimi-K3; access, quotas, and model availability can change.
01
Open a page with the character context
Keep a public character sheet, lore page, or writing brief in a tab. You can reference tabs, screenshots, and files from Tabbit’s input.
02
Choose Kimi-K3 in the model picker
Select Kimi-K3 from the current Chat or new-tab model list. The screenshot below is an interface example, so use the live picker for the latest roster.
03
Ask, compare, and keep reading
Use the sidebar for a focused question, or compare several models in one view. This is a browser chat route, not a replacement for character cards, lorebooks, or ST group chat.
CHOOSE THE FRONTEND
SillyTavern and Tabbit solve different parts
There is no need to pretend one tool has the other tool’s feature set.
| SillyTavern | Tabbit | |
|---|---|---|
| Character cards and lorebooks | Built for this workflow | Use an open page or file as context |
| Kimi K3 connection | Provider, key, endpoint, template | Pick the current built-in model |
| Preset control | Deep control over the RP prompt stack | Chat prompt and Skills, not ST presets |
| Group chat and extensions | Core ST ecosystem | Not the same feature set |
| First reply | More setup, more knobs | Fewer connection steps |
K3 / ST FAQ
Before you spend an evening tuning
What is the official Kimi K3 API model name?+
Moonshot’s official Quickstart uses kimi-k3. Kimi K3 launched through the API on July 16, 2026. Verify the live provider model list because aliases and availability can change.
Is Kimi K3 good for SillyTavern roleplay?+
It can be. Community reports like its prose and dialogue, but also mention overthinking, instruction drift, speed, and 429s. Your card, preset, provider, and context history all affect the result.
Why does Kimi K3 overthink in SillyTavern?+
K3 always has thinking enabled in Moonshot’s API. Try a lower reasoning effort where the provider supports it, keep instructions short, and check whether the frontend preserves the complete assistant thinking history.
Do I need a Kimi K3 preset or prefill?+
You can start without one. If you need tighter RP behavior, look for current community preset or prefill guidance and change one layer at a time. Do not treat a preset as an official Kimi requirement.
Can Tabbit replace SillyTavern?+
No. Tabbit is useful for chatting with Kimi K3 beside a live webpage, but it does not reproduce ST character cards, lorebooks, extensions, or group chat.
Can I use Kimi K3 in Tabbit without my own API key?+
The current Tabbit model picker includes Kimi-K3 as a built-in option. Availability and quotas are product settings that can change, so check the picker after installing.
Try Kimi K3 before tuning another preset
Keep SillyTavern when you need its character stack. When you want the model next to a page, open Tabbit, choose Kimi-K3, and start with the context you already have.
Available for macOS and Windows. Model access and quotas may vary by edition and plan.