Try Kimi at kimi.com →

Kimi K2.6

Kimi K2.6 is the everyday model in the Kimi family. It is the one you chat with for free in the Kimi app, and the low-cost choice for developers. It handles questions, writing, coding and agent tasks, understands pictures, and is the only current Kimi model that can answer instantly or think first, as you choose.

Released 20 April 2026Current model
Kimi K2.6 ONE MODEL, TWO SPEEDS InstantThinking You choose, per task EVERYDAY WORK ChatWriteCode AND FOR BIG JOBS A swarm of up to300 agents The low-cost all-rounder of the Kimi family

Fast when you want speed, careful when you want depth.

We are an independent guide. Kimi K2.6 is made by Moonshot AI. We are not affiliated with them. The scores on this page are Moonshot's own published figures, not our tests. Prices and plans change, so confirm them on Moonshot's sites.
QUICK ANSWER

The short version

  1. 1

    What it is

    Kimi's general-purpose model for everyday chat, writing, coding and agents.

  2. 2

    What it costs

    Free to chat with in the app. $0.95 in and $4 out per million tokens on the API.

  3. 3

    What stands out

    Thinking you can switch off, and Agent Swarm with up to 300 helpers.

THE MODEL

What Kimi K2.6 is

A model is the "brain" behind an AI assistant. Kimi K2.6 is the general-purpose one: good at many things and priced for daily use. Moonshot released it on 20 April 2026 as the follow-up to Kimi K2.5, with the same design and size but better results, especially on long tasks.

1 trillionparameters, with 32 billion active at a time
256Ktokens it can read at once. Roughly 400 pages
Text + imagesas input. Video is experimental
Openweights, under a Modified MIT License

From Moonshot's Kimi K2.6 model card. The release date is from news coverage at launch. The page estimate uses about 650 tokens a page. New to these words? See how Kimi works.

ABILITIES

What Kimi K2.6 is good at

Moonshot's model card highlights four things.

Long coding jobs

It can stay on one programming task for hours. Moonshot describes runs of more than 12 hours and 4,000 tool calls.

Design from a prompt

It turns a short description or a picture into a working interface, with layout, buttons and animation.

Agent Swarm

It can split a big job between up to 300 helper agents over 4,000 steps, then join the results into documents, sites or spreadsheets.

Always-on agents

It is built to run background agents that keep working on a schedule. It is the default model behind Kimi Claw.

See these in use: Agent Swarm in K2.6 · Kimi Claw · Design to Code · Agent Mode

MODES

Instant or thinking: you choose

Kimi K3 and Kimi K2.7 Code always think before they answer. Kimi K2.6 lets you decide. That makes it the natural pick when speed and cost matter more than depth.

Instant mode

  • Answers straight away
  • Uses fewer tokens, so it costs less
  • Best for quick questions, short edits and chat

Thinking mode

  • Works through the problem step by step first
  • Slower, and uses more tokens
  • Best for maths, planning and difficult code

On the API, thinking is on by default and developers switch it off in the request. The modes offered in the Kimi app may be named differently.

SCORES

How it scored

First, the improvement over Kimi K2.5 on six tests. Higher is better.

SWE-bench Verified

Fixing real software bugs

K2.576.8
K2.680.2

SWE-Bench Pro

Harder, longer bug fixes

K2.550.7
K2.658.6

Terminal-Bench 2.0

Coding tasks in a terminal

K2.550.8
K2.666.7

BrowseComp

Finding hard facts online

K2.574.9
K2.683.2

HLE-Full, with tools

Expert-level questions

K2.550.2
K2.654.0

MMMU-Pro

Understanding pictures

K2.578.5
K2.679.4

Second, against the leading models from other companies at the time it launched, in April 2026. The best score in each row is marked.

TestKimi K2.6GPT-5.4Claude Opus 4.6Gemini 3.1 Pro
SWE-Bench ProHarder bug fixes58.657.753.454.2
SWE-bench VerifiedFixing real bugs80.2Not given80.880.6
DeepSearchQADeep web research92.578.691.381.9
HLE-FullExpert questions, no tools34.739.840.044.4
MMMU-ProUnderstanding pictures79.481.273.983.0

On a small screen, swipe the table sideways to see every column.

Read these with care. Every figure comes from Moonshot's model card. Moonshot ran many tests itself with its own setup and took some rival scores from other reports. We did not run any of them. The rivals shown have since been replaced by newer models, and so has Kimi K2.6 at the top of its own family, by Kimi K3.
GET STARTED

Where to use Kimi K2.6

The Kimi app

Everyday chat at kimi.com and in the mobile apps runs on Kimi K2.6, free.

How to use Kimi

Agent features

Agent tasks and Agent Swarm can run on it, using your plan's credits.

Agent Swarm

Kimi Claw

The always-on agent uses Kimi K2.6 unless you switch it.

Kimi Claw

The API

For developers. The model name is kimi-k2.6. Pay per token.

Kimi K2.6 API

Other providers

Because the weights are open, some cloud companies host it too. Prices and limits are theirs.

Your own servers

Download the weights from Hugging Face. Needs a multi-GPU server.

Notes below

That free chat uses Kimi K2.6 comes from our pricing pages. Confirm on kimi.com.

COST

What Kimi K2.6 costs

API, per million tokensInputCached inputOutput
Standard$0.95$0.16$4.00
Batch$0.57$0.10$2.40

On a small screen, swipe the table sideways to see every column.

API prices from Moonshot's price pages, before tax. More: Kimi API pricing · Is Kimi AI free?

FOR DEVELOPERS

Notes for developers

Full walkthrough: Kimi K2.6 API · Kimi API setup

CHOOSE

Kimi K2.6, K2.7 Code or K3?

Kimi K2.6Kimi K2.7 CodeKimi K3
Made forEveryday general workProgrammingThe hardest tasks of any kind
ThinkingOn or offAlways onAlways on
Reads at onceAbout 262K tokensAbout 262K tokensAbout 1 million tokens
API price, in / out$0.95 / $4.00$0.95 / $4.00$3.00 / $15.00
Batch priceYesYesNot listed
Pick it forChat, writing, quick tasks, high volumeDaily codingHuge documents and deep reasoning

On a small screen, swipe the table sideways to see every column.

A simple rule: start with Kimi K2.6. Move up to Kimi K3 only for the jobs where its answers fall short.

More: Kimi K2.7 Code · Kimi K3 · choose a Kimi model

HONEST NOTES

Limits to keep in mind

FAQ

Kimi K2.6 questions

Kimi K2.6 is Moonshot AI's everyday general model, released on 20 April 2026. It handles chat, writing, coding and agent tasks, understands images as well as text, and lets you switch thinking on or off.

Chatting with Kimi K2.6 in the Kimi app is free, according to our pricing pages. On the API it costs $0.95 per million input tokens and $4.00 per million output tokens, before tax.

Kimi K3 is larger, reads about four times as much at once and scores higher on hard tasks. Kimi K2.6 costs about a third as much, and it is the only current Kimi model that can answer without thinking first.

Yes. Moonshot reported 80.2% on SWE-bench Verified. For daily programming, Kimi K2.7 Code is the specialist built on top of it, and it uses fewer thinking tokens.

Moonshot says Agent Swarm with Kimi K2.6 scales to 300 sub-agents and 4,000 coordinated steps, up from 100 and 1,500 with Kimi K2.5.

Its weights are published on Hugging Face under a Modified MIT License. Running it needs powerful server hardware. Read the license before commercial use.
NEXT

Related guides

Try Kimi K2.6 for free

Kimi K2.6 runs on Moonshot AI's own website and apps. We are an independent guide and do not offer sign-in or downloads here.