Skip to content

← All series & guides

Moonshot

Kimi from scratch

0 of 6 parts live

Kimi is Moonshot's assistant, famous for long context. The app, the K2/K3 weights, and what long context actually buys you.

  1. 1 What Kimi is (Moonshot) in plain English Theo uploaded a 410-page due-diligence PDF because Kimi advertises a million tokens, then got a cite from page 12 of a different deal. Kimi is Moonshot AI's assistant (chat app, K2 and K3 weights, and an API), and the window is a suitcase, not memory. Scheduled · September 25, 2026
  2. 2 Kimi Long-Context Windows Without Mythology Two-million-token context windows sound like permanent memory, but attention dilution and retrieval degradation still strike when facts sit in the middle of massive prompts. Here is how Kimi handles long documents and how to verify retrieved facts. Scheduled · October 14, 2026
  3. 3 Your First Hour with Kimi: Docs, Research, and Writing Most users treat long-context chat tools like casual search engines, missing their real power as synthesis engines. Here is how to set up workspaces, frame multi-document extraction prompts, and produce executive research dossiers in your first sixty minutes with Kimi. Scheduled · October 15, 2026
  4. 4 Kimi File Upload Habits and Practical Limits Uploading files into a long-context model feels effortless until truncated tables, corrupted OCR text, and bloated token bills derail your workflow. Here are the disciplined upload habits and hard technical limits every analyst needs to master. Scheduled · October 16, 2026
  5. 5 Hosted Kimi vs Moonshot Open Weights in Plain English Choosing between Moonshot's hosted cloud API and self-hosting their open weights is not just a hardware math puzzle. Here is how latency, enterprise data privacy, token pricing, and operational maintenance stack up in real production environments. Scheduled · October 17, 2026
  6. 6 When Kimi Is the Wrong Tool Long-context synthesis models excel across massive document corpora, but forcing them to perform deterministic math, realtime low-latency chat, or complex codebase refactoring invites disaster. Here is where Kimi fails and which specialized engines to deploy instead. Scheduled · October 18, 2026