Editing Openai/691a14bb-ac48-8008-85d0-2e4fb3f780d1 (section)

===== What it is: =====
* A Mixture-of-Experts (MoE) model by Moonshot AI. 1 trillion total parameters, about 32 billion active per input. Hugging Face<ref>{{cite web|title=Hugging Face|url=https://huggingface.co/moonshotai/Kimi-K2-Instruct|publisher=Hugging Face|access-date=2025-11-17}}</ref>
* Context window up to ~256 K tokens in its “Instruct” version. Together.ai Docs<ref>{{cite web|title=Together.ai Docs|url=https://docs.together.ai/docs/kimi-k2-quickstart|publisher=Together.ai Docs|access-date=2025-11-17}}</ref>
* Optimised for “agentic” workflows: tool-use, reasoning, coding across large tasks. Weights & Biases<ref>{{cite web|title=Weights & Biases|url=https://wandb.ai/site/inference/moonshotai-kimi-k2/|publisher=Weights & Biases|access-date=2025-11-17}}</ref>

Strengths:
* Very large context window: helps when you dump many files, large codebases, or multi-file refactoring.
* Strong coding + reasoning capabilities: suited for complex architecture tasks or tooling chains.
* Good for agent-driven integration (e.g., multi-step workflows) rather than simple snippet generation.

Weaknesses / trade-offs:
* Likely higher latency and cost compared to smaller models.
* Possibly overkill for small tasks (single file, trivial edits).
* May require careful prompt engineering to maximise the MoE benefits.

When to use in Windsurf Cascade:
* Use when you are orchestrating or refactoring a large project: e.g., “analyse entire repo, identify dependencies, propose module restructure”.
* Use when there is deep tool chaining: e.g., you want to auto-generate tests, docs, commit changes across many files, with reasoning about each step.
* Avoid if your task is short / isolated (better use a faster, cheaper model).