Articles

Deep-dive AI and builder content

What Kimi Code CLI / IDE can do: capabilities, cost, IDE integration, and best-fit workflows

As of July 2, 2026, Kimi Code is a terminal AI coding agent. The official README says it can read and edit code, run shell commands, search files, fetch web pages, and continue based on feedback. It also lists video input, MCP configuration, plugins, subagents, hooks, and ACP/IDE integration.

Capability Visible capability Why it matters Guardrail
Read and edit code yes changes the repository directly inspect every diff
Run shell commands yes runs tests, builds, scripts confirm risky commands
Search files yes locates entry points in unfamiliar repos constrain large repositories
Fetch web pages yes pulls docs and issue context verify source quality
MCP / plugins / hooks / subagents yes connects tools and splits work audit permissions
ACP / IDE kimi acp works with Zed, JetBrains, ACP clients login and repo permissions still matter
Entry point Command or action Best for Note
macOS / Linux script install.sh + curl -fsSL <install.sh> \| bash fastest CLI setup official README says no Node.js required
Homebrew brew install kimi-code repeatable install convenient for developer machines
Windows PowerShell install.ps1 + irm <install.ps1> \| iex Windows users install Git for Windows first
First run kimi, then /login OAuth or API key users run inside the project directory
IDE / ACP kimi acp Zed, JetBrains, ACP clients connect after CLI login
Model Context Input cache hit Input cache miss Output Best fit
kimi-k2.7-code 262,144 tokens $0.19 / 1M tokens $0.95 / 1M tokens $4.00 / 1M tokens coding model and repo-agent tasks
kimi-k2.7-code-highspeed 262,144 tokens $0.38 / 1M tokens $1.90 / 1M tokens $8.00 / 1M tokens high-speed coding sessions, about 180 tokens/s and up to about 260 tokens/s in short context
kimi-k2.6 262,144 tokens $0.16 / 1M tokens $0.95 / 1M tokens $4.00 / 1M tokens general multimodal, tool use, JSON mode, and agent work
kimi-k2.5 262,144 tokens $0.10 / 1M tokens $0.60 / 1M tokens $3.00 / 1M tokens lower-cost Kimi K2.x experiments
moonshot-v1-128k 131,072 tokens n/a $2.00 / 1M input tokens $5.00 / 1M output tokens V1 long-context generation reference
deepseek-v4-flash 1M $0.0028 / 1M tokens $0.14 / 1M tokens $0.28 / 1M tokens low-cost API baseline
50k input + 10k output sample Cache miss cost Cache hit cost Read this as
deepseek-v4-flash $0.0098 $0.00294 lowest API cost baseline here
deepseek-v4-pro $0.03045 $0.00888 stronger DeepSeek tier, still cheaper than Kimi K2.x in this sample
kimi-k2.5 $0.0600 $0.0350 cheapest Kimi option in this set
kimi-k2.6 $0.0875 $0.0480 general Kimi multimodal/agent option
kimi-k2.7-code $0.0875 $0.0495 Kimi coding model baseline
kimi-k2.7-code-highspeed $0.1750 $0.0990 higher price for interactive speed

Estimate cost from the underlying model because the README does not publish a separate Kimi Code monthly subscription price. For a 50k input + 10k output coding task, kimi-k2.7-code costs about $0.0875 on cache miss or $0.0495 on cache hit. kimi-k2.6 is the general multimodal/agent comparison point; moonshot-v1-128k is an older V1 reference, not the Kimi Code-first estimate. Token cost is usually smaller than reviewer time.

A good 30-minute pilot starts on a clean branch: ask for a read-only repo explanation, give one README/test/script task, require command output, then review touched files and failure notes. Kimi Code fits small repository edits, codebase explanation, test additions, and IDE-assisted coding. It does not fit direct production-branch changes, secret directories, automatic deploys, or broad refactors without tests.

Official sources: Kimi model list, Kimi K2.7 Code pricing, Kimi K2.6 pricing, Kimi K2.5 pricing, Moonshot V1 pricing, Kimi Code GitHub, DeepSeek models pricing, DeepSeek API docs.

Related reading

RadarAI helps builders track AI updates, compare source-backed signals, and decide which changes are worth acting on.

← Back to Articles