Field manuals
Guides for builders
Fast answers first, operational details second, and primary sources attached to every changing claim.
Model selection · 6 min
Choose K3, K2.7 Code, or HighSpeed
Match model capability to workload depth, visual input, execution length, latency, and quota.
API · 5 minSend a first K3 request
Use the current model ID, official endpoint, and a minimal OpenAI-compatible client.
Kimi Code · 7 minInstall and configure Kimi Code
Get the current CLI, authenticate, select a model, and avoid the cache trap during comparisons.
Migration · 4 minMove off retired K2 model IDs
Inventory hard-coded names, retest tool behavior, and plan around the next documented retirement date.
Cost · 4 minDesign for K3 context caching
Why stable prompt prefixes and session boundaries can matter more than the headline input price.
Verification · 3 minRead the current signal
See what Kimi Insider has verified, what remains pending, and where every claim came from.