Chrome ExtensionKimi K3.1 Pro Prompt Generator
Updated for Kimi K3.1 Pro (October 2026), Moonshot AI's 3T-class open-weight model with always-on thinking, text, image and video input, and a 1M-token context. Pick Chat, Agent or Agent Swarm, set the reasoning effort and task horizon, and get a clean brief that sets clear limits on what Kimi should and should not do.
Be specific about what you want the AI to do
Generated Prompt
Fill in the form and click "Generate" to create an optimized Kimi K3.1 Pro prompt.
Tip: The more specific your task description and context, the better the generated prompt will perform.
Kimi K3.1 Pro Tips
- • K3.1 Pro always thinks. There is no Instant mode, so skip "think step by step" and pick a reasoning effort instead
- • Kimi can be overly proactive. State what it must NOT do: files to leave alone, scope limits, when to stop and ask
- • For long-horizon coding, name the repo structure, coding standards and the exact success criteria
- • Agent Swarm: describe the whole job and how it splits into parallel pieces, then say how the results should be merged
- • The 1M-token context takes whole codebases, long documents and video. Put the material first and your instruction last
- • Attach images or video directly and say exactly what to look for and what to return
Kimi K3.1 Pro Prompt Templates
Copy-ready starting points for the most common jobs people bring to this model. Swap the [BRACKETED] parts for your own details, or use them as a shape to imitate when you write your own.
Agent Swarm Task
AgenticKimi decomposes long-horizon work across sub-agents — name the decomposition.
Goal: [THE END STATE] Available tools: web search, file system, code execution, sub-agents. Suggested decomposition: 1. [SUBTASK] — research 2. [SUBTASK] — implementation 3. [SUBTASK] — verification Constraints: [CONSTRAINTS]. Stop and ask if [STOP CONDITION]. Report the plan first, then execute, then summarise what each sub-agent produced and how you reconciled conflicts.
Whole-Document Analysis
AnalysisUses Kimi's very large context — long material first, question after.
[PASTE THE FULL DOCUMENT SET] --- Based on everything above, [THE TASK]. Requirements: - Cite the document and section behind each point. - Surface contradictions between documents rather than smoothing them over. - Note explicitly if any part of the input seems truncated or unreadable. Format: [FORMAT].
Multi-File Feature Build
EngineeringLong-horizon coding, which is Kimi's strongest mode.
## Goal Build [FEATURE] end to end. ## Stack [LANGUAGE + VERSION], [FRAMEWORK], [DATABASE], [TEST FRAMEWORK]. Conventions: [LINTER/STYLE]. ## Existing code [PASTE RELEVANT FILES OR REPO TREE] ## Requirements [NUMBERED FUNCTIONAL REQUIREMENTS] ## Constraints No new dependencies. Preserve [EXISTING BEHAVIOUR]. Migrations must be reversible. ## Output Format Plan → each new/changed file in full → migration steps → tests → how to verify locally.
Reasoning Mode Problem
ReasoningFramed so K3's always-on thinking has something worth deliberating over; raise reasoning effort for harder problems.
Problem: [THE PROBLEM] Known constraints: [CONSTRAINTS] Known unknowns: [WHAT YOU DON'T KNOW] Work through this carefully, checking each edge case, then give the final answer. Before finalising, double-check the result against the constraints above and state whether it satisfies all of them. If it cannot, say which constraint has to give.
Research Report
ResearchSearch-and-synthesise work with a defined evidence standard.
Research [TOPIC] and produce a report for [AUDIENCE]. Scope: [WHAT'S IN, WHAT'S OUT]. Timeframe: [RECENCY REQUIREMENT]. Requirements: - Cite every factual claim with a source and date. - Distinguish primary sources from commentary. - Include a "what the sources disagree about" section. - Flag where the evidence is thin rather than filling the gap with plausible prose. Format: executive summary, findings, disagreements, gaps, sources. [WORD COUNT] words.
Document to Structured Output
DataLow-effort shaped: tight, answer-shaped, no preamble.
Extract [WHAT] from the document below into [FORMAT, e.g. a Markdown table with columns X, Y, Z]. Rules: source values verbatim, blank cell where absent, no invented entries, no commentary before or after the output. Document: [PASTE DOCUMENT]
Kimi Model & Prompting Update Log
K2 series retirement
● LatestOctober 31, 2026- Since October 31, 2026, the K2 series (including K2.6 and K2.7 Code) is no longer offered to new API users. Existing users keep access for now.
- Moonshot plans to retire all K2 models by the end of December 2026 and recommends migrating to Kimi K3.1 Pro.
Kimi K3.1 Pro open weights
November 4, 2026- Moonshot published the full Kimi K3.1 Pro weights on Hugging Face (moonshotai/Kimi-K3.1-Pro), keeping its promise to ship them by November 4.
- The model card lists 2.8T total parameters with 104B active, a 1M-token context, and MXFP4 weights trained with quantization-aware training.
- It ships under Moonshot's own Kimi K3.1 Pro License rather than a standard open-source licence.
Kimi K3.1 Pro
October 23, 2026- Moonshot released Kimi K3.1 Pro, its most capable 3T-class open-weight model, post-trained from K3. API IDs are kimi-k3.1-pro and kimi-k3.1-pro-highspeed (a faster variant), and it is selectable in Kimi Code.
- It keeps the 1M-token context and always-on thinking, with reasoning effort low, high or max (the API default is max). Temperature and top_p stay fixed at 1.0 and 0.95.
- The API takes text, images (base64 or an uploaded ms:// file) and video uploads, returns up to 1,048,576 output tokens (131,072 by default), and supports strict JSON schema output, partial mode and context caching with a 5-minute or 1-hour TTL.
- Moonshot reports far fewer hallucinations than K3, so it is more willing to say it does not know. Prompts that ask it to flag uncertainty benefit most.
- K3 is now the previous-generation flagship and still available.
Kimi K3 on Amazon Bedrock
September 18, 2026- Kimi K3 became generally available on Amazon Bedrock, with native vision and the full 1M-token context.
- AWS says it is the first open-weight model on Bedrock to support explicit prompt caching, which cuts cost and latency when you reuse the same long context across calls.
Kimi K3 open weights
July 27, 2026- Moonshot released the full K3 weights under its own Kimi K3 License, not a standard open-source licence.
- Per reporting, the licence asks inference providers earning more than $20 million a year from the model to share up to 30% of that revenue.
- At 2.8T total parameters it is the largest open-weight model available.
Kimi K3
July 16, 2026- A 2.8T-parameter sparse Mixture-of-Experts model that routes each token to 16 of 896 experts, with about 104B parameters active. It is built on Kimi Delta Attention combined with gated MLA.
- Native text, image and video input in a single model, with a 1,048,576-token context window.
- Thinking is always on. Depth is set with reasoning_effort (low, high or max, default max); there is no separate non-thinking or Instant model as in earlier Kimi generations.
- Temperature and top_p are fixed on the API (1.0 and 0.95), so steering has to happen in the prompt.
- In multi-turn and agent use, the full assistant message including its reasoning must be sent back in the conversation history.
- Debuted at #3 on the Artificial Analysis Intelligence Index with a score of 57, behind Claude Fable 5 and GPT-5.6 Sol.
- Agent Swarm on kimi.com now runs on K3, with up to 300 sub-agents and more than 4,000 tool calls in a single task.
Kimi K2.x
January - June 2026- K2.5 (January 2026): 1T total parameters with 32B active, multimodal, with separate Instant and Thinking modes.
- K2.6 (April 2026) and the K2.7 Code model (June 2026) brought a 256K context and stronger long-context coding.
How to Use the Prompt Generator
Define Your Task
Select a category and describe what you want Kimi to do. Choose a mode (Chat for a single answer, Agent for multi-tool workflows, Agent Swarm for large parallel jobs) and a reasoning effort.
Add Context & Constraints
Provide background context, set your desired tone and output format, and specify any constraints. For coding tasks, include language version and framework details.
Generate & Use
Click "Generate" to get an optimized prompt tailored to your selected Kimi mode. Copy and paste it into kimi.com, Kimi Code, the Kimi API (kimi-k3.1-pro) or Amazon Bedrock.
Frequently Asked Questions
What is Kimi K3?
Kimi K3 is Moonshot AI's previous-generation flagship (succeeded by K3.1 Pro in October 2026), released on July 16, 2026, with open weights following on July 27. It is a 2.8-trillion-parameter Mixture-of-Experts model, the largest open-weight model available, with about 104B parameters active per token. It takes text, images and video as input and has a 1,048,576-token context window. On release it placed third on the Artificial Analysis Intelligence Index, behind Claude Fable 5 and GPT-5.6 Sol.
What is new in Kimi K3?
Three changes affect how you prompt it. First, thinking is always on: the Instant and Thinking split from K2.5 is gone, and you choose a reasoning effort of low, high or max instead (max is the API default). Second, video joins text and images as native input, and the context grows from 256K on K2.6 to 1M tokens. Third, Moonshot notes that K3 can be overly proactive in agent work, so prompts should say plainly what it must not change and when it should stop and ask. It is also available on Amazon Bedrock since September 18, 2026.
Which Kimi mode should I pick?
Chat is for a single answer or deliverable. Agent is for multi-step work with tools such as web search, a file system or a code interpreter, where the model plans and runs the steps itself. Agent Swarm, which runs on K3 in the Kimi app for paid members, splits a large job across up to 300 parallel sub-agents, which suits wide research, batch document processing and big coding projects. Reasoning effort is a separate choice: use Low for quick, well-defined tasks and keep Max for the hardest problems.
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is Moonshot AI's newest flagship, released on October 23, 2026. It is a 3T-class open-weight model post-trained from Kimi K3, with a 1M-token context, image and video input and always-on thinking. On the API it is kimi-k3.1-pro, with a faster kimi-k3.1-pro-highspeed variant. Moonshot reports a big drop in hallucinations compared with K3, and Moonshot published the full weights on Hugging Face in early November 2026 (2.8T total parameters, 104B active).
How does this prompt generator work?
You provide structured inputs: task, role, context, mode, reasoning effort and task horizon. The generator applies what works on Kimi K3.1 Pro: a clean task spec with no "think step by step" padding, explicit limits on scope and actions, a tool list and decomposition for Agent and Agent Swarm work, and long material placed before the instruction.
Is this tool free to use?
Yes. You get 1 free prompt generation per day with no signup required. For unlimited access, sign up for a Promptslove membership which includes all AI tools and 29,000+ premium prompts.
What is Kimi best at?
Kimi K3.1 Pro is strongest at long-horizon coding (multi-file refactors, end-to-end features), agentic workflows with long chains of tool calls, and vision-grounded work such as turning screenshots or video into code or analysis. Its 1M-token context also suits whole-codebase and whole-document analysis.
Can I use these prompts with other AI tools?
While optimized for Kimi, the generated prompts follow universal principles and work well with ChatGPT, Claude, Gemini, MiniMax, Qwen, DeepSeek, and Mistral. Mode-specific instructions (Chat / Agent / Agent Swarm) are Kimi-specific but degrade gracefully on other platforms.
Is Kimi free to use?
The Kimi app at kimi.com has a free tier with usage limits; Agent Swarm needs a paid membership. On the API, kimi-k3 is billed per token: $3.00 per million input tokens on a cache miss, $0.30 on a cache hit, and $15 per million output tokens. Because K3 always thinks, its reasoning tokens are billed as output, so a lower reasoning effort can noticeably cut cost. For kimi-k3.1-pro, price trackers list $2 per million input tokens ($0.20 on a cache hit) and $10 per million output tokens, with kimi-k3.1-pro-highspeed at double that; batch jobs bill at 60% of the standard price. Check Moonshot's pricing page for current rates.
How many prompts can I generate for free?
You get 1 free prompt generation every 24 hours. This resets automatically. For unlimited generations, consider signing up for a Promptslove membership.
Want Unlimited AI Prompt Generation?
Get unlimited access to all AI tools, 29,000+ premium prompts, courses, and resources designed to maximize your creative output.