Kimi Insider
A field desk for the models, agents, APIs, releases, and real-world changes shaping the Kimi ecosystem.
The Signal Board
What changed, what is current, and what builders need to verify before they ship.
Kimi K3 redraws the model map with a 1M-token context window
Moonshot AI has moved its flagship to a 2.8-trillion-parameter, native-vision model built for long-horizon coding, knowledge work, and reasoning.
The Wire
All dispatchesKimi Code now has two distinct lanes: K3 and K2.7 Code
The coding product now separates frontier reasoning from a focused coding model and a high-speed variant.
APIK3 API pricing makes cache behavior part of the architecture
The headline prices are $0.30 cached input, $3 cache-miss input, and $15 output per million tokens.
WatchThe K2 migration window is closing faster than the model names suggest
The original K2 API series is already retired, while K2.5 and Moonshot V1 access changes are now tied to the K3 rollout.
ReleaseK2.6 made long-horizon coding the center of Kimi's story
The April release established the coding, design, and agent-swarm direction that K3 now extends.
Current Model Map
Open the reference1,048,576 tokensLong-horizon coding, knowledge work, multimodal reasoningCurrentKimi K2.7 CodeFocused coding262,144 tokensReliable daily coding and repository workCurrentK2.7 Code HighSpeedFast coding262,144 tokensLow-latency coding loops; higher quota useCurrentKimi K2.6General multimodal262,144 tokensFast chat, visual tasks, and existing K2.6 workflowsSupportedField Guides
Browse all guidesChoose K3 or K2.7 Code without guessing
A workload-first decision guide for reasoning depth, coding focus, speed, and context.
API quickstartMake a first K3 request
The endpoint, model ID, reasoning settings, and a minimal OpenAI-compatible example.
Kimi CodeInstall, sign in, and pick a model
Official install commands plus the practical difference between K3 and the K2.7 coding lanes.