Overview
kimi-k2.6 is Moonshot’s newest agentic flagship (April 2026). It handles tool calls, long sessions, and image-aware workflows more naturally than the rest of the catalog. Beats K2.5 on internal agentic evals per Moonshot’s release notes. This is the model the agent and best-agent aliases resolve to.
Specs
| Field | Value |
|---|---|
| Model ID | kimi-k2.6 |
| Best for | Agentic coding, tool use, multimodal workflows |
| Context window | 262,144 tokens |
| Max output tokens | 32,768 |
| Input modalities | Text, image |
| Output modalities | Text |
| Tool calling | Yes |
| Structured outputs | Yes |
| Prompt caching | No |
| Speed | Medium |
| Cost band | Premium |
| Release stage | Stable |
Pricing
| Per 1M tokens | |
|---|---|
| Input | $1.28 |
| Output | $5.40 |
Use this when
- Your coding agent uses lots of tool calls.
- You need long context for large repos or multi-step tasks.
- You want one model for text plus screenshots or diagrams.
- You care more about reliable agent behavior than lowest latency.
Pick something else when
- You want the best general-purpose default: use
qwen-3.6-plus. - You only need deep reasoning, not agent behavior: use
deepseek-r1. - You need a cheaper long-context option: use
gemini-2.5-flash. - You need the previous-gen agentic model: use
kimi-k2.5.
Example
agent alias to always get the latest agentic flagship:
Agent query example
Related aliases
| Alias | Resolves to |
|---|---|
agent | kimi-k2.6 |
best-agent | kimi-k2.6 |