Updated

A small brushed-silver crescent reflected in a dark still pool on a dark background, abstract hero artwork for Kimi K2.6

Moonshot AI

Kimi K2.6

Moonshot AI's open coding model that turns prompts and visual inputs into working interfaces.

Chat with Kimi K2.6 in Super AgentView pricing

What is Kimi K2.6?

Kimi K2.6 is an open-weight model from Moonshot AI, released in April 2026, with 1T total parameters and 32B active.

Moonshot built it for long coding tasks in languages such as Rust, Go and Python, for coding-driven design that turns prompts and visual inputs into production-ready interfaces, and for Agent Swarm runs of up to 300 sub-agents and 4,000 coordinated steps. It reads text and images in a 256K-token context window, and Moonshot's own API also accepts video as an experimental feature. In Super Agent its thinking is a simple on or off switch, on by default.

Provider
Moonshot AI
Context window
256K tokens
Input
Text, Image
Reasoning
On or off
Released
April 2026

How Kimi K2.6 performs

Published by Moonshot AI

Results Moonshot AI reported for Kimi K2.6. Each lab runs its own setup, so compare these with other models only inside the same table.

Benchmark results Moonshot AI published for Kimi K2.6
BenchmarkScoreSetting
Humanity's Last Exam (full)54.0%with tools, thinking mode
BrowseComp83.2%thinking mode
SWE-Bench Verified80.2%thinking mode
SWE-Bench Pro58.6%thinking mode
Terminal-Bench 2.066.7%Terminus-2 harness, thinking mode
LiveCodeBench v689.6%thinking mode
AIME 202696.4%thinking mode
GPQA Diamond90.5%thinking mode
OSWorld-Verified73.1%thinking mode
MMMU-Pro79.4%thinking mode

Source: huggingface.co

What you can do with Kimi K2.6

  • Turns prompts and visual inputs, such as screenshots, into working interfaces

  • Long coding tasks in Rust, Go and Python

  • Agent Swarm of up to 300 sub-agents, according to Moonshot

  • Open weights under a modified MIT license

  • Simple on or off thinking

How to prompt Kimi K2.6

  • Attach a screenshot and ask it to describe or rebuild the layout.
  • Leave reasoning on for planning; switch it off for quick replies.
  • Name the language or tool you work in so its code matches your setup.

Where Kimi K2.6 fits best

  • Turning a screenshot into a layout description

  • Code-heavy creative tools

  • Multi-step agent work

  • Interface and landing page ideas

How Kimi K2.6 compares

Kimi K2.6 compared with other chat models on Flashloop
ModelLMArena textLMArena codingContext windowReasoning
Kimi K2.61461#591516#44256K tokensOn or off
Claude Sonnet 51462#581520#371M tokensOff to max
GPT-6 Sol1459#621519#381.05M tokensOff to max
DeepSeek V4 Pro1458#641501#791M tokensOff, medium, high, extra high
Grok 4.61456#661510#60500K tokensLow to extra high

Good to know before you start

  • 256K-token context, smaller than the 1M of Kimi K3
  • No graded effort levels, only on or off
  • Interfaces come out as code; for images and video, Super Agent uses Flashloop's generation models

Start chatting with Kimi K2.6

Chat with Kimi K2.6 in Super Agent

  1. Select Kimi K2.6

    Open Super Agent on Flashloop and choose Kimi K2.6 in the model picker.

  2. Enter your prompt

    Ask for what you need: an answer, a plan, or the images and videos you want the agent to make.

  3. Send and keep going

    Send the message. Kimi K2.6 answers in Super Agent and runs any generations you asked for.

Questions about Kimi K2.6

What is Kimi K2.6 best at?

Long coding tasks, turning prompts and visual inputs into interfaces, and large multi-agent runs. Moonshot says its Agent Swarm can coordinate up to 300 sub-agents.

What inputs does Kimi K2.6 accept?

Text and images, with a 256K-token context window. Moonshot's own API also takes video, which it labels experimental.

How do I control reasoning on Kimi K2.6?

It has an on or off switch in Super Agent, and it starts with reasoning on.

Kimi K2.6 or Kimi K3?

Kimi K2.6 focuses on coding and interface work and has a 256K window. Kimi K3, Moonshot's current flagship, adds a 1M-token context and effort levels, and video is built into the model. On K2.6, video input is experimental and limited to Moonshot's API.

More chat models on Flashloop

Browse all models